Latest / Science Behind the Glass: illuminating the human side / Unveiling the Code: AI in Science with Raiyan Talkhani
Transcript
- 0:01Who are the characters behind science? This is
- 0:04Science Behind the Glass, illuminating the human
- 0:06side. In this podcast, we unveil the human side
- 0:11of laboratories and science. Scientists share
- 0:14personal experiences and we discuss their meanings
- 0:17from different perspectives and fields of research.
- 0:25Welcome to Science Behind the Glass, illuminating
- 0:28the human side. Today we are diving into the
- 0:31world of artificial intelligence, a realm that's
- 0:34both revered and feared. In the scientific arena,
- 0:38AI has become an indispensable tool, revolutionizing
- 0:42fields like medicine, conservation, and modeling.
- 0:45Our special guest today is Ryan Takhani, a skilled
- 0:49software engineer who harnesses the power of
- 0:52AI for image processing tasks. Ryan holds a bachelor
- 0:57in electrical and computer system engineering
- 1:00from Monash, Malaysia. and a Master of Engineering
- 1:04of Internet of Things and Data Engineering from
- 1:08James Cook University, Cairns, Australia. His
- 1:11expertise shines through AI projects where he
- 1:15delved into the face detection and rural road
- 1:19extraction from satellite images. But that's
- 1:22not all. Ryan also is the co -director of Quizland
- 1:26AI. former director and also have an startup
- 1:31and we are going to delve more in his experience
- 1:34and his prospects. Ryan, welcome to Science Behind
- 1:38the Glass. Thanks for joining us today and to
- 1:41start this episode. could you please explain
- 1:44a little bit about your background a little bit
- 1:47about you oh thank you so much yeah uh you explained
- 1:51it quite well uh from my career point of point
- 1:54of view it was uh electrical and computer systems
- 1:56engineering at monash university and like funnily
- 2:00enough that's where i first got my taste of ai
- 2:02i made a project for them back in 2017 for my
- 2:06t -scenes which was uh identifying faces even
- 2:09when half of the face was covered or part of
- 2:12the face was covered So this was back then when
- 2:14AI couldn't detect them. So that was my project.
- 2:18So that is where I got my first ever taste of
- 2:20AI. Yeah, and this was in Malaysia. So I grew
- 2:24up in Saudi Arabia. So as soon as I finished
- 2:27my schooling in Saudi Arabia, I moved to Malaysia
- 2:29for my university. I wanted to do IT, but it
- 2:34was back then electrical and computer systems.
- 2:38Engineering was offering more opportunities.
- 2:41So I did that. At that moment, I went back to
- 2:44India. I did my internship in electrical engineering.
- 2:47And then I started looking into other opportunities.
- 2:50And I was looking all over the place. Canada,
- 2:53Malaysia again. And I landed in Australia. And
- 2:58I think by far the best choice out of the options
- 3:00I had at the time. But yeah, I got to do my Internet
- 3:06of Things and Data Engineering Master's or that.
- 3:10That is where I got really into AI, because I
- 3:13had my master's thesis in this field, where I
- 3:17had to create an AI that could look at satellite
- 3:21images on rural and remote roads, identify which
- 3:25ones are not basically on the Google map, and
- 3:29just draw it out. So that was a lot of the math
- 3:32they are missing out on rural and remote roads,
- 3:34right? So to bridge this gap, they wanted an
- 3:37AI that could do that. Right now, it was Jayden,
- 3:40who was manually drawing out roads for his entire
- 3:44thesis. And he did such a good job as well. Jayden,
- 3:47they even had some volunteers, but mostly Jayden
- 3:52and Yoko. And they managed to map out over 2
- 3:56million kilometers of road manually by hand.
- 4:00And it was by manually looking through Google.
- 4:05Google Earth, Satellite Cube, and Matt, yeah.
- 4:08So they didn't even know where to, yeah, so they
- 4:10identified where to look for it with their research.
- 4:13And for this research, right? Wow, that sounds
- 4:14painful. It is. You know how it is, like collecting
- 4:18data properly. So this is where the first ever,
- 4:22my first ever approach in like how AI can be
- 4:26used for science came into being. Yeah. Can you
- 4:30explain us a little bit, what is the Internet
- 4:32of Things? Internet of Things. Alright, back
- 4:35in 2020, 2019, there was a big hype around these
- 4:40devices. So, Internet of Things is basically
- 4:44smart connected devices. And they do have one
- 4:47smart component of it. For example, if you have
- 4:51the DAB connected to internet, that is just a
- 4:55massive product connected to internet. That doesn't
- 4:58mean anything. But Internet of Things is basically...
- 5:02machine to machine communication which is if
- 5:05you had a fridge which could look at what is
- 5:08there in your fridge and update your grocery
- 5:10list now that is internet of things it is machines
- 5:14communicate uh communicating with different components
- 5:16or different devices and components within itself
- 5:20or outside itself looking at what data is given
- 5:24to them and updating their data or updating uh
- 5:28making decisions based on the data they have
- 5:30given so the fact is So that was Internet of
- 5:34Things. And this is, again, comes in really handy
- 5:38in sites, especially who are going out in the
- 5:41field, deploying sensors, monitoring different
- 5:43regions. They don't just need data to be collected.
- 5:47They need decisions to be made by the device.
- 5:50Technology of internet device is monitoring,
- 5:53for example, controlling the glucose or the levels
- 5:57of sugar in blood and telling the person... communicating
- 6:01with the phone of the person and telling like,
- 6:03oh, it's too high or too low. That would be an
- 6:06example? That would be an example of IoT device.
- 6:09Yeah, that's a really good example. Oh, that's
- 6:12great. And can you explain us in simple words,
- 6:15like for general public, what is AI? Oh, AI is
- 6:19artificial intelligence. People can look at things,
- 6:23they can make decisions, they can understand
- 6:25like when my hands show up, right? For me. It
- 6:29is like, oh, this is my hand, this is me. But
- 6:32for computers, it is just colors of pixels. It
- 6:35doesn't mean much to them. So up until a couple
- 6:39of years ago, the majority of the computers and
- 6:42the majority of the computer software was where
- 6:47people were manually hard -coding the programs,
- 6:52manually writing out all the instructions which
- 6:55the computer has to perform in certain different
- 6:57positions. They had to manually identify what
- 7:00they needed to do, and they had to encode all
- 7:03of the scenarios. Now, a couple of years back,
- 7:06artificial intelligence came in. Now, this was
- 7:08where computers could actually make decisions
- 7:10based on the data. They could interpret data
- 7:14the way people could, to a certain extent. Now,
- 7:17they did that by, instead of manually coding
- 7:20everything right, all they did was, okay, computer,
- 7:24here is the data. Here is what. what I'm studying.
- 7:30Here are the parameters I'm looking at. And these
- 7:33are the responses which we need for certain values
- 7:37of these parameters. So you give the computer
- 7:40a list of those parameters or values. And once
- 7:45you give back to the computer, you have already
- 7:48programmed the computer to look at the data and
- 7:51learn a certain way. So instead of manually encoding
- 7:55the responses, or manually encoding the response
- 7:58data you are actually programming the computer
- 8:02to learn from the data and adjust its decision
- 8:07making so once the computer learns right it can
- 8:10then look at the data and it can refer back to
- 8:13its training or its learnings instead of referring
- 8:16back to what the person has coded AI is really
- 8:21important for scientific work because the AI
- 8:24can help analyze data at a larger scale A lot
- 8:27of scientists write when, if it is an individual
- 8:29person or a team of people that are looking at
- 8:32data, they can only look at certain small scope
- 8:35of data before they make the generalizations,
- 8:37before they make their statements, or before
- 8:39they do their research. Regardless of how big
- 8:43the team is, you are only capable of looking
- 8:46at limited amount of data. Now what happens is
- 8:51all these researchers, all these people who hire
- 8:53as data analysts are They certainly have to,
- 8:57you need to pay them. You need to fund the research.
- 9:00You need to get the resources for them. So it
- 9:03costs a lot of money. Now, AI can do a lot of
- 9:07these, well -trained AI can do a lot of these
- 9:10processes of what people take days or months
- 9:13to do. It can do it in minutes or hours. It can
- 9:17process a large amount of data based on how many
- 9:19resources you give it. And it can process the
- 9:22data and come to a conclusion or make. uh to
- 9:24an analysis of data within a matter of minutes
- 9:27instead of days or weeks and this is and it can
- 9:32do the it can do the analysis and it can work
- 9:36as if it is working in a team of people or a
- 9:39group of people so it has so much power when
- 9:43it is plus a simple data ai is really good tool
- 9:46for researchers to use because whatever data
- 9:49they collect they can progress faster within
- 9:52their research something which a person a master's
- 9:56student does his entire thesis on now could be
- 10:00a scale of a one subject or one means of work
- 10:03for a student or a researcher so you you see
- 10:07like how fast the progress in science will be
- 10:09if they're allowed to process a higher scale
- 10:12data at a larger scale or large capacity so it's
- 10:16it's optimizing the all the process to science,
- 10:24for example. Yes, and not just optimize, but
- 10:27also processing it at the largest speed at which
- 10:32processes are made and done, and conclusions.
- 10:35Kill and conclusion. Because it can copy the...
- 10:39This is where I know it is controversial to say.
- 10:42To a certain degree, it can copy the decision
- 10:45-making skills of a skilled individual based
- 10:48on the data you provided to a certain extent.
- 10:51So this is where having that type of skill within
- 10:54your research is going to be really important
- 10:56for scientists than just using computers. That
- 11:01sounds exciting. And what ignited your belief
- 11:05in the pivotal role of AI in science? What drew
- 11:10you into this fascinating convergence of technology
- 11:13and research or other potential field works?
- 11:18Oh, okay. So for me, my background was in engineering,
- 11:21but I've always been a fan of science and always
- 11:23been a fan of research. I always enjoyed looking
- 11:26at what other people are doing, other research
- 11:29people are doing in biology, marine biology.
- 11:33And I wanted to be more involved in these types
- 11:37of research. So that's what got me into working
- 11:42with AI for researchers. Now, the other thing
- 11:45with AI, what... The researchers I worked with,
- 11:48they either didn't have data or they had really
- 11:51limited data. So how do we work with that? So
- 11:55that was the other fascinating thing, the other
- 11:57challenge AI is facing now. It's basically, how
- 12:01do you work when you have a limited amount of
- 12:03data? And how can you train AI models? How can
- 12:09you make inference with this? So that really
- 12:11fascinated me as a researcher, and that got me
- 12:14involved in the scientific process. because they
- 12:16are always at the edge of understanding of what
- 12:19they believe about certain things, what they
- 12:21believe about certain medicine, certain concepts,
- 12:23and how they can change their belief. And being
- 12:27there with my tools present for the researchers
- 12:30to use, that was a different kind of challenge
- 12:33which I really found. Yeah, it was like I was
- 12:36assisting people make important discoveries in
- 12:39their field. So that's what brought me. Yes,
- 12:42and especially when you, for example, as you
- 12:45said, having enough data to get some conclusions
- 12:51from the data. Sometimes we don't have enough
- 12:55number of animals or sample size. And using AI
- 12:58helps to draw more insightful conclusions from
- 13:03that without going and getting more animals or
- 13:06if there are not more. We can process the data
- 13:09in a different way. That's amazing. And now,
- 13:12scientists might not be engineers, but AI provides
- 13:16tools that can optimize scientific endeavors
- 13:19across different fields. How can scientists leverage
- 13:23these tools without in -depth and specialized
- 13:26knowledge of an engineer? Oh, okay. I think 2023
- 13:30has been a really good year for AI -related tools.
- 13:35What has happened is that strategy, BTE, Bing,
- 13:38brings so many other systems, especially large
- 13:43language markets, which are generated with AI
- 13:44search engine, which uses GPT -4. I think scientists
- 13:49and researchers, there's already tools which
- 13:53are easy to use, which are easy to understand,
- 13:56which are quick to adapt, which people can quickly
- 14:00adapt. And the user interface is truly easy for
- 14:04these students. And scientists can start looking
- 14:06at using these tools to further understand what
- 14:10other people are doing with their research. So
- 14:12one of the most important thing which comes when
- 14:15it comes to any of the research, it is using
- 14:18AI for doing literature review. So this used
- 14:22to be something which took six months or six
- 14:24months, 11 months. Now this is something that
- 14:27can be done within seconds using GPD or using
- 14:32Yeah, using Bing's GPT -4 because what it does
- 14:35is it goes through all of the search results
- 14:37and Google's bar, you feed any document into
- 14:41it, it goes through it and it understands what
- 14:43all you can do with it. It understands what's
- 14:46happening, it summarizes it for you, and it looks
- 14:48at what other people are doing in similar fields.
- 14:51Now, because I have been enrolled in other search
- 14:55fields, I need to be doing search equals in different
- 14:58fields. which i'm not an expert in but i still
- 15:01need to come to meetings without meetings uh
- 15:04with the knowledge of what each of the researchers
- 15:07are doing so linda if i come and have a meeting
- 15:11with you i'd really understand what your meeting
- 15:13is about because you are doing something related
- 15:17to drug discovery but oh and then the next day
- 15:21i have to meet somebody in the field of dentistry
- 15:25so they have different challenges they have different
- 15:27should really take challenges. Now these are
- 15:29two completely different things. How do I prepare
- 15:31for it? I use AI to help me understand the research
- 15:35papers in this field, problem statements in this
- 15:37field, and because AI is trained on so many specialized
- 15:42knowledge, we need specialized tools and so many
- 15:45specialized knowledge, it is also really good
- 15:47at dumbing it down for me to understand. So whenever
- 15:51I complete video meetings, right? I go through
- 15:53the research papers, but I also have a co -pilot,
- 15:56big co -pilot, right next to me, which I use
- 15:59to explain certain concepts within those research
- 16:02papers or certain concepts within them. So I
- 16:06think if you, for example, give kids an AI to
- 16:10work on, honestly, do it on supervision, but
- 16:12if you give kids an AI to work, it will be the
- 16:17same as scientists because they are going to
- 16:19ask. questions which not all adults can answer.
- 16:23Why is the sky blue? Why is the ocean blue? So
- 16:27for that case, if you look at it like a literature
- 16:30review where you're looking at research what
- 16:32other people have already been doing it, I think
- 16:35that would be a more patient and more easier
- 16:37way for a kid to understand, same as adults,
- 16:40to understand difficult thoughts. So what are
- 16:44the risks that we have to be mindful? of integrating
- 16:47AI into the scientific world. To be honest, it's
- 16:51unique. AI is a really good tool. AI is really
- 16:54simple to use and you don't need any skills.
- 16:59Even the user interface that I'm working on,
- 17:01you don't really need much in the way of understanding
- 17:05how AI works. You can actually understand how
- 17:09AI works, right, basically. Now, the thing is
- 17:13that you shouldn't... worry about ai as a tool
- 17:16but you need to worry about who is deploying
- 17:19that tool because ai is only trained on producing
- 17:23a certain result right but you also need to give
- 17:27it feed it something for it to produce a search
- 17:30results it could be something in your private
- 17:33life it could be some secret discovery or reaching
- 17:36it could be something uh yeah so it could be
- 17:39something which is for example, proprietary code,
- 17:43proprietary gene coding, or like something which
- 17:46are trade secrets, or you might not want to enter
- 17:50it into an AI platform because AI is consistently
- 17:54interlinked. AI is not a set -and -go sort of
- 17:57thing. You need to constantly improve your AI
- 17:59methods. So when you do that, right, when you
- 18:03actually do this, AI is going to be, you need
- 18:05to be more careful about where you're uploading
- 18:08your scientific data. That would be the number
- 18:10one risk because it could be used for training
- 18:12the next iteration of AI and that could result
- 18:16in somebody else being rediscovered by just entering
- 18:19a couple of forms rather than, yeah, so that
- 18:22could happen. That is one of the risks. So it's
- 18:24like a misleading information or depending on
- 18:26who is feeding the AI or? Yes, yes. What other
- 18:31people can do, whether your data is being used
- 18:35for training data, that is one of the risks you
- 18:37need to look into. The other risk, you need to
- 18:39look into is the ai is there's a lot of privacy
- 18:43concerns with ai so again this is only going
- 18:47to come with where you are uploading your data
- 18:51this is one of the problems i had when i was
- 18:53building my general purpose ai platform i'll
- 18:55just be posting it in the background soon uh
- 18:57so the only way i could mitigate this problem
- 19:00for researchers and scientists was okay you can
- 19:03train your own ai models with this but once you
- 19:06delete the data even i won't have a copy of the
- 19:08data so uh so that is yes exactly i don't own
- 19:14any of that data i don't own any of this your
- 19:16post processing network it is yours once it is
- 19:19done it is done it is not going outside of my
- 19:22privacy uh outside outside the so so that is
- 19:26one way of one risk with the ai uh number two
- 19:30the risk is over reliance on ai now there's a
- 19:34new uh the new tools generated by is are really
- 19:37good at researchers doing like i said literature
- 19:39right but this is one skill which researchers
- 19:43and students need to have knowing how to do literature,
- 19:48knowing how to look at research papers, understanding
- 19:50the data. You can use AI to help you understand
- 19:53the data better or help improve your skills of
- 19:56using AI to understand your research papers.
- 19:59But there shouldn't be an over -reliance on AI
- 20:01where you just forget certain skills, forget
- 20:04how to do certain things. And that is going to
- 20:06be another risk for researchers. The over -reliance
- 20:10on AI, that you forget certain skills. See, I
- 20:16told you it was a controversial topic. I know.
- 20:20Handle it properly. Use it properly. It can be
- 20:22amazing. But misusing it and believing all that
- 20:25shit. Yeah. And the third most important thing
- 20:30is biases. Now, the risk of AI, right? Yeah,
- 20:33it is a good tool and it overaligns its power.
- 20:36But also, it is a tool which is trained on data
- 20:39which has already been fed to it. Now, that data
- 20:41can be... bias now if the data which it has been
- 20:45trained on is biased right all of your research
- 20:48all of your analytics are going to be inclined
- 20:51towards a certain will lean towards a certain
- 20:53direction than the other direction now it is
- 20:56all going to be based on the storage data and
- 20:59if historically people have been done things
- 21:01which are wrong or which are bad it is going
- 21:04to be 10 times more prevalent in the research
- 21:08or it what i'm referring to is more like ai not
- 21:13being able to recognize people of different skin
- 21:16color. For example, if I was brown or if I was
- 21:19black, the initial AI mode couldn't actually
- 21:22recognize it because in the training data, they
- 21:24only used white people's photos to train the
- 21:27AI mode. So that is a massive bias. So when the
- 21:31AI ethics people looked into this, they discovered
- 21:35that AI was really not good at finding people
- 21:39of black or brown skin color. That was number
- 21:42one. Again, twice based on data. The other one
- 21:46was Amazon at one point used AI to hire people
- 21:50based on their skills. But the AI only ended
- 21:53up hiring men because it was, again, trained
- 21:57on historic data. So the men's resumes were under
- 22:02no priority historically over women's resumes,
- 22:06right? So because of that bias, storage bias,
- 22:10it also went into AI. And then it started, yeah,
- 22:14it started, like, eliminating all the women from
- 22:18the hiring pool. So they mitigate this issue
- 22:21by removing the addresses, removing names, removing,
- 22:24like, certain things which were prominent, any
- 22:26markers which identified as women, and they used
- 22:30to interact with me. That was one way. It's like
- 22:33a running experiment. You have to try to avoid
- 22:35bias. and randomize everything that we don't
- 22:39have exactly in the results exactly yeah and
- 22:44then we are also curious to learn more about
- 22:47your startup and your ideas could you give us
- 22:51a glimpse into your business and how your intersect
- 22:55how this intersects with the fascinating world
- 22:58of artificial intelligence so my goal with startup
- 23:01was to create specialized rooms for researchers
- 23:04to you with their own data and with the trade
- 23:08models which are specialized to be used with
- 23:11their own data. So that's where I got an idea
- 23:17for the startup and I looked into GCU for example.
- 23:21There were a lot of researchers in here. There
- 23:24are a lot of people who are collecting data at
- 23:26large scale. There are a lot of people who are
- 23:29going into hiking for example. they're going
- 23:33underwater, they're going into tropical forests
- 23:35and they're collecting data. Now I, with my background
- 23:38in AI, I wanted to find ways in which I could
- 23:42help them develop this, help them further understand
- 23:47more about my data, help them understand more
- 23:49about the tools they have on the internet, and
- 23:53help make AI for them to process their own data.
- 23:56So that's what my startup was all about to most
- 23:58researchers. it was again my passion my hobby
- 24:01working with researchers working with businessmen
- 24:03identifying what they can do with their data
- 24:05and giving them the tools necessary to work with
- 24:08that data oh that's amazing because giving a
- 24:11little bit of background to people that doesn't
- 24:13know that one of the biggest steps in science
- 24:16is analyzing data yes it's difficult it's not
- 24:19easy not everybody has the same skills so for
- 24:22example there are no statisticians and it doesn't
- 24:24mean that they just work with a statistician
- 24:27and they understand everything So that's a big,
- 24:30big process that they have to go through understanding
- 24:33what type of analysis they are going to carry
- 24:36out, what is meaningful for the collection of
- 24:40the data, what analysis can or can't not be done.
- 24:44So that's a big step. And I think that is amazing
- 24:48what you are doing because you help not only
- 24:51to run the... basic analysis of what the researcher
- 24:55is thinking about also to suggesting other type
- 24:58of analysis that can do or potentially highlight
- 25:02a result that get from those data that is collected
- 25:07that's some ways that's so oh thank you and to
- 25:10add to that like my specialty was in images so
- 25:13a lot of these a lot of these researchers right
- 25:16they have a lot of images that have like thousands
- 25:21and thousands of frames which need to be analyzed
- 25:24and in this case they have to be analyzed with
- 25:26an individual person to manually look through
- 25:29it and then label it as whether it is this or
- 25:32whether it is that so they have to constantly
- 25:34manually do that to what i was doing was like
- 25:37okay i'll make a tool that can process thousands
- 25:40and thousands of images within a matter of minutes
- 25:43rather than yeah rather than having students
- 25:46manually go through it having research yeah and
- 25:51it takes time it just time consuming having a
- 25:53phd student manually go through all that data
- 25:55it will take me and considering that they have
- 25:58a phd scholarship for three years for example
- 26:02with ai you still need to do some quality analysis
- 26:05quality control it's not like self -driving cars
- 26:08just here you still need to have somebody behind
- 26:11the wheel that's what what is important to highlight
- 26:13that ai is more like that the tool is like they
- 26:16are going to replace humans actually it's a tool
- 26:19it's a it is a tool a lot of people there's a
- 26:23lot of scare going on right now where people
- 26:26are talking about ai is going to replace their
- 26:29jobs ai is going to straight up cancel industries
- 26:33and to be honest ai might AI will do that. Let's
- 26:38be really honest here. AI is going to make changes
- 26:41and it's going to be jobs which are going to
- 26:44be replaced by computers. But this is also going
- 26:48to give us something which we haven't been looking
- 26:51towards for a long time. Instead of doing tasks
- 26:55which are repetitive or instead of doing things
- 26:59which are, which is basically the same thing
- 27:01you are doing over and over again, you can actually
- 27:03automate computers right now. now as an as a
- 27:07person instead of sitting in the office all day
- 27:10instead of sitting in the lab all day you can
- 27:13because you have an ai processing all of this
- 27:16for you you can spend more time with your family
- 27:18and spend more time with your people oh with
- 27:21your folks your friends you can have more time
- 27:23to yourself and you can have more time to think
- 27:26and analyze the results rather than processing
- 27:29the data wondering what you're going to get out
- 27:32of that data now is this going to replace job
- 27:34yes it is going to it is going to take new jobs
- 27:38but but we are only scared because we cannot
- 27:41see in the future how the new jobs are going
- 27:43to evolve now for example you have if you have
- 27:47the data analysts are going to be replaced by
- 27:50consultants for example now they they will be
- 27:54able to consult in jobs as well because they
- 27:56can analyze the data and give feedback from the
- 27:59person consultants job definitely replaced because
- 28:02AI can do the job better So you don't really
- 28:05have to hire consultants, expensive consultants
- 28:08to work with certain, for example, people like
- 28:15them. You don't need to hire consultants to learn
- 28:18about it, to analyze it, to make decisions. You
- 28:20just give AI the data and it can do the consulting
- 28:26for you. Now the consultants, on the other hand,
- 28:28their jobs will evolve into... being business
- 28:31managers, being business assistants, they can
- 28:34take more of the management roles of the people
- 28:36so they can actually take on more responsibility
- 28:40because if their jobs are replaced, they can
- 28:42move into a higher team. They will have to know
- 28:45how to control those tools. Otherwise, adapt
- 28:48or... Yeah, for sure. That sounds a little scary,
- 28:52but to be honest, this is what happens to Steve
- 28:54and Jill. This is what happens. I think... it
- 28:57more yeah it's a tool that we are developing
- 29:00now we we have to be able to manage and to use
- 29:05it properly and yeah i don't know and i leverage
- 29:08and always skills yeah yeah and i see like that
- 29:14yeah yeah no but to be honest like a lot of jobs
- 29:21which ai is going to replace you need to ask
- 29:24yourself are these really the jobs we want to
- 29:26do in the first place it's repetitive it is like
- 29:30something people are being exploited for yeah
- 29:33that's true and like it takes a lot of time for
- 29:35example report that if i ai could help for example
- 29:39doctors and with the reports doing the report
- 29:42better spending hours doing administrative work
- 29:45instead of investing that time doing research
- 29:48or checking and the patients so i think and managing
- 29:52your startup alongside your work in ai sounds
- 29:57challenging and do you do you manage your own
- 30:00time like uh how do you balance your personal
- 30:03life and oh because like for me at the moment
- 30:08over here it has only been for uh because this
- 30:13has been my life for a couple of months now and
- 30:16i genuinely enjoy it i genuinely enjoyed working
- 30:19on that i used to go for heights uh once in a
- 30:22while i used to go for yeah i got earlier Yeah,
- 30:26I used to go outside to meet friends more often
- 30:28than I do now. But being an entrepreneur and
- 30:31working as a researcher is a very challenging
- 30:34thing. You need to be really passionate about
- 30:37what you do. You need to be really into what
- 30:39you do because you are going to be working with
- 30:42a lot of difficult data, really a lot of difficult
- 30:46data, a lot of difficult systems. And it will
- 30:49not just be in difficult systems for AI architecture.
- 30:53but also understanding what people are doing
- 30:55in areas of that kind of difficulty. So you won't
- 30:59really have much time to do other things. If
- 31:01you are, it's not something you can just turn
- 31:04off. You have to be on it. What is your opinion
- 31:07about the concerns of these technologies and
- 31:11the reliance of them to solve social and environmental
- 31:15issues? Social and environmental issues. Okay,
- 31:19now this is a really good question. yeah because
- 31:23let's start with environmental issues now because
- 31:25i know about this using ai is going to be offset
- 31:29it leaves a lot of carbon waste because it it
- 31:34will cost you the whole year because here's the
- 31:37thing when you're training ai right you need
- 31:39to use a lot of gpus and using gpus overheats
- 31:42a lot it takes up a lot of energy so a lot of
- 31:45data centers or you will have in between training
- 31:49uh treating air mountains and using self -driving
- 31:52cars uh using electric cars and everything i
- 31:55think the consumption of electricity is going
- 31:58to go up drastically in the next couple of years
- 32:01because intensive computers use ai is really
- 32:04expensive to run so uh i mean it is cheap compared
- 32:07to hiring but it is expensive in a sense that
- 32:10it is that that's a lot of waste that decimal
- 32:14use that's the motor and it takes up a lot of
- 32:16energy and using and energies are created using
- 32:20coal, energy is created using renewable resources.
- 32:23It is going to have multiple impact on the planet
- 32:26and with our drastic increase in the use of AI,
- 32:29we are also drastically increasing the use of
- 32:31electrons. So it is going to have major environmental
- 32:33impact. And also to make those computers, to
- 32:37make those computers, to make those data centers,
- 32:40it is going to require a lot of resources which
- 32:43are sourced, some not in any good way. No, mining.
- 32:47Yeah. mining, but yeah, a lot of unethical ways.
- 32:51So you need to, AI is going to have a lot of
- 32:54challenges over time as well because certainly
- 32:56there's such a need for these resources and we
- 32:59need to find, we need to claw out the right way
- 33:02to meet these requirements. And this is not something
- 33:06you can use to solve social problems, but it
- 33:08is going to cause a lot of social problems, because
- 33:10no matter where you are, there will be controversies
- 33:13on how you train an AI to respond to a certain
- 33:16question, which will be controversial. which
- 33:18could be controversial in a certain week of capitalism.
- 33:21Now, because of the data, it's streamed on the
- 33:23brain. So for social issues, this is going to
- 33:27have a hundred. It's going to be more challenging
- 33:29because whatever the AI is echoing, people might
- 33:33echo. And vice versa, whatever people are echoing,
- 33:37AI will echo the loudest. Because of AI, because
- 33:41of biased streaming data, because it is going
- 33:43to be down to get to a certain level of analytics.
- 33:47what is in your professional personal life oh
- 33:51yes i will be moving to brisbane main goal before
- 33:55i start my job in brisbane is to host my ai platform
- 34:00and just leave it running in the background so
- 34:02i'll be giving you a link for debugging and testing
- 34:05and debugging stage andrew will be debugging
- 34:07it andrew will be handling everything so yeah
- 34:11so so you're moving to brisbane for a company
- 34:16Yes, I got, because of all the consulting I did
- 34:19for my company over here, they require the skills
- 34:22investment as well. They need something to actually
- 34:25understand how the data acts in the world, how
- 34:28they can actually, how they can really develop
- 34:30AI related work for people in different forms.
- 34:33So this is what I've been doing for a year. I
- 34:35used to meet you, I used to talk to researchers
- 34:37and everyone, right? And I used to... to help
- 34:40them understand how AI solution can be created,
- 34:43created, made using official products or creating
- 34:47a product for them. This was something I already
- 34:49did. So to help with this, this game is something
- 34:54which they need for the startup. Again, I'm working
- 34:58for. So they need this. It's called Co -Pilot
- 35:02HQ. They are solutions architectures. What they
- 35:04do is they look at how people are currently working
- 35:07and they find ways to implement AI within them.
- 35:09That sounds cool. Congratulations. Thank you.
- 35:12Thank you. Like, I'm really excited. I'm really
- 35:14excited about the handovers when my project will
- 35:17go live. And it will be handled, everything will
- 35:20be handled by Andrew. So, yeah, I'll just be.
- 35:24Andrew is your partner in the company, right?
- 35:29Andrew, going to be. It's going to be. Going
- 35:32to be. Because, yeah, I need, because this is
- 35:35not something I can manage on my own, right?
- 35:38Yeah, no. Yes. That's great. And lastly, tell
- 35:43us an anecdote that you have had during your
- 35:46studies or your career paths. And it can be funny
- 35:50or it can be a challenging one or both. I would
- 35:53say my most favorite moment. It is interesting
- 35:57or it is very defining for me as my career as
- 36:01an AI developer as well. It was back when I was
- 36:04doing my research with Jaden, right? So one of
- 36:08the things that happened was, My AI was not working.
- 36:11It was worrying for the submission of my thesis.
- 36:14My AI was just not performing. I couldn't get
- 36:16it to reach the levels or reach the accuracy
- 36:21levels which I needed for it to actually, by
- 36:25individual strength, I could not get it to work.
- 36:27And I was panicking and I was panicking. And
- 36:30there's this really smart lecture I had. I needed
- 36:36somebody who didn't know anything about my research
- 36:39so I couldn't go to my supervisor at the time
- 36:42as well like well I needed somebody with a fresh
- 36:44perspective so I randomly go to this genius I
- 36:50would say he's having poor guys trying to enjoy
- 36:53his you know engineering part of you you have
- 36:56like one every semester where all the engineers
- 36:57come up the students are you so I just go to
- 37:01him and I'm having my workout and I'm like a
- 37:04high random question i know you're enjoying your
- 37:09burger but i'm having so so so problems my ai
- 37:12is not reaching the accuracy level what is going
- 37:15wrong with it and then he gives me any like just
- 37:19without hesitation the guy is so so good at the
- 37:23time he just points out like okay do this this
- 37:27your your models are not reaching like a certain
- 37:31threshold your training is insufficient this
- 37:33and it gives me a list of things that are wrong
- 37:35and one week for submission that basically two
- 37:37days later like my training was like i stream
- 37:40it all night anyway it performed twice as well
- 37:44it just it just happened dr ronson phillip yeah
- 37:49but my supervises were also really helpful but
- 37:51this is not something which we could figure out
- 37:53at the time why it had happened but and i think
- 37:57uh this list is very important because those
- 37:59are some things which i still i still like playing
- 38:03with this and i have yeah and i have wanted to
- 38:07have not my understanding of ai standing on how
- 38:09to train ai models based on the conversation
- 38:11i had with them what an amazing interview i personally
- 38:21like ai as a tool and use it almost daily for
- 38:24a repetitive task However, it needs to be used
- 38:28wisely and not blindly accept everything it provides
- 38:31or suggests. I make it a point to double check
- 38:34citations and concepts whenever I use it. What
- 38:37are your thoughts about Dani? Yes, those are
- 38:40good precautions. This is a controversial hot
- 38:43topic right now, and as usual, it seems to be
- 38:45about taking sides, pro -AI and anti -AI. I think
- 38:49it's interesting to address the legitimate concerns
- 38:51of people more on the anti -AI side, because
- 38:54there are important discussions that are still
- 38:56scarce. For example, the role of governmental
- 38:59regulation and the importance of building consensus
- 39:01around the limitations that this technology might
- 39:04need. Nevertheless, this is already happening.
- 39:06And I think there is no turning back. So it is
- 39:08urgent to have these discussions at every level
- 39:10of society. I agree. The ethical concerns need
- 39:13to be addressed. I think that as it evolves,
- 39:16it's also necessary to guarantee people that
- 39:18it is safe to use. And as Ryan pointed out, depending
- 39:22on the task or field, who is feeding the system
- 39:25needs to be careful with information that provides
- 39:27to the AI, because this can lead to biases and
- 39:31discrimination. Yes, that is right. In the end,
- 39:33I see that this technology, as many others, will
- 39:36probably reflect the creators as they are, for
- 39:39good or for bad. I think we as humans are at
- 39:41a breaking point where we need to take a good
- 39:43self -evaluation and decide to be better, because
- 39:46this technology will probably co -evolve with
- 39:48us. Absolutely. It is important to recognize
- 39:50that just because AI is effective on one specific
- 39:53task doesn't necessarily mean it can be seamlessly
- 39:56applied to other tasks or fields. Training the
- 39:59tool is crucial. And it's vital to approach AI
- 40:03with a realistic mindset, steering clear the
- 40:06misconception that it can solve all problems
- 40:08like a fortune teller. Each application requires
- 40:11careful consideration and thoughtful implementation
- 40:14for optimal results. Right. Something similar
- 40:18has happened with different technologies. Maybe
- 40:20in this case, it seems a bit easier to use these
- 40:23chatbots than using, for example, the first computers.
- 40:26So it might be easier to misusing it or misinterpreting
- 40:29it. So I think that may be the obstacles that
- 40:33scientists can find on using this kind of technology.
- 40:37And with this, we reach the end of this episode.
- 40:39We thank you for listening and our interviews
- 40:42for their time. Tune in again in two weeks for
- 40:45more Science Behind the Glass, Illuminating the
- 40:47Human Side. This episode was inspired by curiosity,
- 40:50personal experiences, and the paper Artificial
- 40:53Intelligence in Science, an Emerging General
- 40:55Method of Invention. We will leave the references
- 40:58to these materials in the episode description.
- 41:01This podcast is self -funded and self -edited
- 41:04by its host. Find us on our YouTube channel,
- 41:08Spotify, and our official website, Science Behind
- 41:11the Glass. Follow us in our social media on X,
- 41:14Facebook, and Instagram. Original music by our
- 41:17colleague and friend, Bernardo Aguililla.