Latest / Azadeh and Guy on AI / Can AI Build a Better World?

Transcript

Can AI Build a Better World? · Azadeh and Guy on AI

• 0:00
  SPEAKER_02: Hi and welcome to Azadev and Guy on AI.
• 0:04
  SPEAKER_00: And this is episode 10 and we are super excited to have Omid back with us.
• 0:11
  SPEAKER_01: Thank you so much.
• 0:11
  SPEAKER_02: It's the second time we are recording with Omid.
• 0:14
  SPEAKER_00: Yes.
• 0:15
  SPEAKER_00: He is now friend of the pod.
• 0:20
  SPEAKER_00: We should give an award of some kind.
• 0:23
  SPEAKER_00: Anyway, friend of the pod.
• 0:25
  SPEAKER_00: He had a lot to talk about AI and we decided to just have a second installment
• 0:31
  SPEAKER_00: and really talk about how he integrates AI into his work.
• 0:35
  SPEAKER_00: We talked a lot about how he integrated AI into school and all the ups and downs of that.
• 0:42
  SPEAKER_00: But now we're interested in how he's working.
• 0:45
  SPEAKER_02: With AI, to find different job opportunities.
• 0:48
  SPEAKER_02: We also have some very specific questions about the research that he's doing.
• 0:54
  SPEAKER_02: And Omid has published a lot.
• 0:57
  SPEAKER_02: And he's also on the job market.
• 0:59
  SPEAKER_02: All right.
• 0:59
  SPEAKER_00: So if you're looking for somebody to hire, I recommend... To all recruiters.
• 1:05
  SPEAKER_02: Hire Omid.
• 1:07
  SPEAKER_00: Let's start with some of your papers that focus on using AI with GIS
• 1:13
  SPEAKER_00: and remote sensing and all of that.
• 1:15
  SPEAKER_00: So tell us a little bit and give us an example.
• 1:19
  SPEAKER_01: So one of the most important things in international development is creating impact.
• 1:28
  SPEAKER_01: And creating impact comes after effective policy.
• 1:35
  SPEAKER_01: So...
• 1:38
  SPEAKER_01: And how we are doing effective policy.
• 1:41
  SPEAKER_01: We are making policies effective when
• 1:47
  SPEAKER_01: they come after research-driven
• 1:51
  SPEAKER_01: data or information.
• 1:53
  SPEAKER_01: So that's the cycle.
• 1:55
  SPEAKER_01: And specifically on the integration of AI, GIS, and international development.
• 2:03
  SPEAKER_01: So that interaction is...
• 2:08
  SPEAKER_01: Ensuring that you're getting access to quality data because AI itself is not in
• 2:14
  SPEAKER_01: the field to make that data.
• 2:17
  SPEAKER_01: So that's the tool that it creates insights from that quality data
• 2:24
  SPEAKER_01: that you have already provided.
• 2:27
  SPEAKER_00: All right.
• 2:28
  SPEAKER_00: So give us an example of something you've done.
• 2:30
  SPEAKER_01: So as far as Azadeh's research and your own research lab is focused on education,
• 2:38
  SPEAKER_01: I can bring an example of the education because education is also a significant pool
• 2:45
  SPEAKER_01: in international development.
• 2:48
  SPEAKER_01: So let's take the example of
• 2:54
  SPEAKER_01: So,
• 3:03
  SPEAKER_01: mostly these less development countries,
• 3:08
  SPEAKER_01: though I'm not sure that Nepal is a less developed country or not,
• 3:12
  SPEAKER_01: but in these countries,
• 3:16
  SPEAKER_01: When there are projects of building schools,
• 3:19
  SPEAKER_01: all these educational infrastructure,
• 3:22
  SPEAKER_01: the whole project is sometimes getting corrupted.
• 3:25
  SPEAKER_01: For example,
• 3:26
  SPEAKER_01: you're living in the village A and you're living in the village B and the distance
• 3:31
  SPEAKER_01: between two villages is like Let's say 50 kilometers,
• 3:35
  SPEAKER_01: but there are three schools within your village and there is none in your village.
• 3:40
  SPEAKER_01: And because there is some political will to make that schools in your area as far
• 3:46
  SPEAKER_01: as there is some strong ties between the religious leaders or village leaders to
• 3:56
  SPEAKER_01: the government or to the decision body.
• 3:59
  SPEAKER_01: So that's the thing.
• 4:02
  SPEAKER_01: Schools are not building based on the data.
• 4:07
  SPEAKER_01: That's not data-driven, that project.
• 4:10
  SPEAKER_01: So what we're trying to do with AI and GIS is that first we are going
• 4:16
  SPEAKER_01: to use satellite or remote sensing.
• 4:20
  SPEAKER_01: The broad term is remote sensing.
• 4:23
  SPEAKER_01: Seeing what's happening in the field.
• 4:27
  SPEAKER_01: We see the roads,
• 4:28
  SPEAKER_01: we see the masks,
• 4:29
  SPEAKER_01: the hospitals, the alleys and the infrastructure and GIS is getting the
• 4:38
  SPEAKER_01: Longitude and the coordinations to map it, to see how this village likes.
• 4:46
  SPEAKER_01: And then,
• 4:48
  SPEAKER_01: so we have the remote sensing data that we have this road here,
• 4:54
  SPEAKER_01: that road there, hospitals, but no school or less schools.
• 4:58
  SPEAKER_01: And Jay is mapping there.
• 5:00
  SPEAKER_01: And we also have some administrative data, for example, the livelihood, the salaries.
• 5:07
  SPEAKER_01: The local businesses,
• 5:08
  SPEAKER_01: the poverty line,
• 5:10
  SPEAKER_01: access to healthcare,
• 5:12
  SPEAKER_01: enrollment records from the Ministry of Education or that Education Authority.
• 5:18
  SPEAKER_01: So now we are unifying all the data sets and now AI step in.
• 5:24
  SPEAKER_01: It says, it drives conclusions.
• 5:28
  SPEAKER_01: It brings insights that within the village X,
• 5:33
  SPEAKER_01: that there is a long distance between the village and the school and
• 5:38
  SPEAKER_01: the roads are destroyed due to conflicts.
• 5:42
  SPEAKER_02: Yeah.
• 5:43
  SPEAKER_01: School enrollment is less than 5%.
• 5:47
  SPEAKER_01: So it is bringing the insight and here we need some professionals to validate
• 5:53
  SPEAKER_01: that spotting trend from the data by the AI and
• 5:58
  SPEAKER_01: then checking how we can take actions.
• 6:02
  SPEAKER_01: As the policy makers.
• 6:03
  SPEAKER_01: So this is the whole process that we can use AI,
• 6:07
  SPEAKER_01: GIS to see how, in what extent we can have impactful development interventions.
• 6:16
  SPEAKER_02: And just because of AI, all of that matters, all of that happens in a matter of seconds, right?
• 6:22
  SPEAKER_02: Very fast.
• 6:22
  SPEAKER_01: Yeah, because if you don't have that quality data,
• 6:27
  SPEAKER_01: for example,
• 6:28
  SPEAKER_01: by remote sensing,
• 6:29
  SPEAKER_01: now when I say remote sensing,
• 6:33
  SPEAKER_01: that industry itself is also getting a lot more developed by AI.
• 6:40
  SPEAKER_01: Within a way that I may not understand it,
• 6:44
  SPEAKER_01: how they are doing, but every sector is getting developed by the AI.
• 6:48
  SPEAKER_00: Yeah.
• 6:49
  SPEAKER_01: So now you're not doing, for example, baseline and end line surveys.
• 6:55
  SPEAKER_01: Mm-hmm.
• 6:56
  SPEAKER_01: That takes ages in a village.
• 6:59
  SPEAKER_01: You're doing imaginary, satellite imaginary, then you're getting quality data.
• 7:05
  SPEAKER_01: But at the same time,
• 7:06
  SPEAKER_01: you have AI that if you are a professional yourself,
• 7:11
  SPEAKER_01: you can use AI efficiently to make that insights valid.
• 7:16
  SPEAKER_01: Because let's say, let's have another example.
• 7:21
  SPEAKER_01: Let's say that we got that dataset ready and brought to the research lab
• 7:28
  SPEAKER_01: and you're giving it to one of your incoming masters, this degree student.
• 7:33
  SPEAKER_01: Hey, you can go through this dataset.
• 7:37
  SPEAKER_01: I'm happy that you joined my lab.
• 7:38
  SPEAKER_01: This is the dataset that you're going to work and you're going to use it
• 7:41
  SPEAKER_01: for your master's thesis.
• 7:43
  SPEAKER_01: So go play with the data.
• 7:46
  SPEAKER_01: And spot some trends.
• 7:49
  SPEAKER_01: So that's the toughest part because when you're playing around with the data,
• 7:54
  SPEAKER_01: you might not come with a solid topic or title.
• 7:58
  SPEAKER_01: But I'm sure that AI is doing that one in the best way because it checks all
• 8:05
  SPEAKER_01: the variables together.
• 8:08
  SPEAKER_01: It goes online to see what are the trends.
• 8:12
  SPEAKER_01: What are the trends in that field?
• 8:14
  SPEAKER_01: It looks all these branches at the same time,
• 8:19
  SPEAKER_01: reflecting the quality of data to the trends and then picking the variables.
• 8:26
  SPEAKER_01: So this one can be your dependent variable.
• 8:29
  SPEAKER_01: These are the explanatory variables.
• 8:31
  SPEAKER_01: You can come up with this idea.
• 8:33
  SPEAKER_01: So that's the thing that AI can be very effective in.
• 8:38
  SPEAKER_02: Yeah.
• 8:39
  SPEAKER_02: Yeah.
• 8:40
  SPEAKER_00: That goes back to the reason we invented statistics,
• 8:44
  SPEAKER_00: actually,
• 8:45
  SPEAKER_00: in many ways, and the way we think about why do we need to summarize everything.
• 8:51
  SPEAKER_00: And partially it is because our brains are so limited that we can hold,
• 8:56
  SPEAKER_00: especially in working memory and that whole cognitive side,
• 9:01
  SPEAKER_00: we can hold a few numbers in our brain.
• 9:03
  SPEAKER_00: We cannot hold a whole spreadsheet with Even dozens of numbers, we already get lost.
• 9:11
  SPEAKER_00: We can cling to this cell or that cell,
• 9:13
  SPEAKER_00: but to see trends and all of that, we have to have summary statistics.
• 9:18
  SPEAKER_00: And this is where AI steps in and can see the whole picture and can spot trends
• 9:23
  SPEAKER_00: and can spot data that may be suspect and all of these things
• 9:28
  SPEAKER_00: in really effective ways,
• 9:30
  SPEAKER_00: especially when you're doing what,
• 9:33
  SPEAKER_00: you know, Azaleh knows this well, but I love data snooping.
• 9:37
  SPEAKER_00: Right.
• 9:37
  SPEAKER_00: Some people call it data analytics.
• 9:39
  SPEAKER_00: The people from statistics say, well, we're kind of stacking the odds here.
• 9:43
  SPEAKER_00: But we all do that to a certain degree.
• 9:46
  SPEAKER_00: It's like, what is the shape of the data?
• 9:49
  SPEAKER_00: What are the trends?
• 9:49
  SPEAKER_00: What are we seeing here?
• 9:51
  SPEAKER_00: And AI, you're right.
• 9:53
  SPEAKER_00: I think absolutely, especially as we get more data, right?
• 9:58
  SPEAKER_00: As you're talking about the satellite data or other data like that,
• 10:02
  SPEAKER_00: it's an enormous amount of data.
• 10:04
  SPEAKER_00: There's no other way to process it, really, without AI.
• 10:07
  SPEAKER_00: As you're saying, even the satellites are using AI.
• 10:10
  SPEAKER_00: Everybody's using AI five times by the time it gets to us as a data set.
• 10:15
  SPEAKER_00: And we just add one more layer of AI, although eventually we have to verify.
• 10:20
  SPEAKER_00: And as you say,
• 10:21
  SPEAKER_00: think about what are the policy implications,
• 10:23
  SPEAKER_00: think about the suggestions and the real patterns that are happening there.
• 10:29
  SPEAKER_02: Let me see.
• 10:29
  SPEAKER_02: I also have another question.
• 10:31
  SPEAKER_02: So one of the hot topics about AI is its environmental impacts and people have different ideas.
• 10:37
  SPEAKER_02: Guy has his own opinions about that.
• 10:41
  SPEAKER_02: So one of your research themes is green AI or making AI itself more sustainable.
• 10:48
  SPEAKER_02: As AI models become larger and they become more energy intensive,
• 10:53
  SPEAKER_02: how do we balance innovation with environmental responsibility?
• 11:00
  SPEAKER_01: That's a technical question.
• 11:04
  SPEAKER_01: It goes beyond my expertise.
• 11:09
  SPEAKER_00: So far.
• 11:10
  SPEAKER_01: Yeah.
• 11:12
  SPEAKER_01: And it's an engineering facet.
• 11:19
  SPEAKER_01: But I would like to come up with a new perspective that...
• 11:27
  SPEAKER_01: These days when there is something new pops up,
• 11:31
  SPEAKER_01: for example, AI,
• 11:32
  SPEAKER_01: there is always some people,
• 11:34
  SPEAKER_01: as Guy said,
• 11:36
  SPEAKER_01: there are some people that are stepping in and saying, no,
• 11:41
  SPEAKER_01: I'm not doing this, I'm not accepting this, this is harming X, this is harming Z.
• 11:47
  SPEAKER_01: There's always these people saying this.
• 11:51
  SPEAKER_01: That's kind of a constant element there.
• 11:54
  SPEAKER_01: But what I'm saying when we're talking about the sustainability and also the AI is
• 12:02
  SPEAKER_01: in what extent it's bringing environmental hazards.
• 12:09
  SPEAKER_02: Okay.
• 12:10
  SPEAKER_01: So we should quantify that.
• 12:13
  SPEAKER_01: That's the thing.
• 12:15
  SPEAKER_01: For example,
• 12:15
  SPEAKER_01: if you are renting out,
• 12:20
  SPEAKER_01: let's say,
• 12:21
  SPEAKER_01: 20 hectares,
• 12:23
  SPEAKER_01: I have no idea of these hectares in English,
• 12:27
  SPEAKER_01: but I think it's going to be huge, right?
• 12:30
  SPEAKER_01: It's going to be big, 20 hectares, right?
• 12:33
  SPEAKER_01: So I imagine that it's a big area.
• 12:36
  SPEAKER_01: So let's say we are renting out 20 acres of
• 12:43
  SPEAKER_01: pasture from Nebraska to a data center
• 12:46
  SPEAKER_01: for, let's say, Anthropik.
• 12:51
  SPEAKER_01: If that model or if that AI is coming up with solutions
• 12:57
  SPEAKER_01: that it gonna contribute Tremendously to the water infrastructure,
• 13:02
  SPEAKER_01: sustainable water infrastructure in Nebraska,
• 13:06
  SPEAKER_01: and that social benefit is outweighing these 20 acres of pasture land to
• 13:13
  SPEAKER_01: a business land, then you are fine with it.
• 13:17
  SPEAKER_01: But it doesn't make sense in my idea that we should block the way of innovation in
• 13:25
  SPEAKER_01: the AI industry because it's hurting.
• 13:28
  SPEAKER_01: Yeah.
• 13:32
  SPEAKER_01: Food and transportation, that's the biggest sectors that they're emitting greenhouse gases.
• 13:38
  SPEAKER_01: But we are not stopping eating.
• 13:41
  SPEAKER_01: We are trying to change the manufacturing of foods or processing of food
• 13:47
  SPEAKER_01: or producing of food in a sustainable way.
• 13:50
  SPEAKER_01: So we are not erasing the concept itself that we should stop
• 13:57
  SPEAKER_01: But we are trying to eat sustainably.
• 14:01
  SPEAKER_01: And this pattern is, I think, kind of true about the innovation, about the AI.
• 14:10
  SPEAKER_01: So the larger the models are,
• 14:13
  SPEAKER_01: the more infrastructure goes for the cooling,
• 14:15
  SPEAKER_01: for the GPUs and all this energy that they need to process their data.
• 14:23
  SPEAKER_01: But at the same time,
• 14:26
  SPEAKER_01: We should advocate,
• 14:29
  SPEAKER_01: and this is a very cliche word,
• 14:31
  SPEAKER_01: but we should raise our voices to be heard that we should quantify
• 14:39
  SPEAKER_01: the social outcome of that AI and if that's outweighing the
• 14:48
  SPEAKER_01: The sustainable investment that we are doing, that's good.
• 14:52
  SPEAKER_01: Also, there is one more point that I want to point out.
• 14:57
  SPEAKER_01: So within the larger models of AI,
• 15:01
  SPEAKER_01: it doesn't necessarily It gives you the accuracy of information for everything.
• 15:07
  SPEAKER_01: For example,
• 15:08
  SPEAKER_01: you might have a very small data set to use AI,
• 15:12
  SPEAKER_01: and you have at the same time like 10 types of AI models.
• 15:19
  SPEAKER_01: When users, I mean,
• 15:21
  SPEAKER_01: not individual users like me,
• 15:25
  SPEAKER_01: but when you're talking about,
• 15:26
  SPEAKER_01: for example,
• 15:27
  SPEAKER_01: a big company,
• 15:29
  SPEAKER_01: they have these big subscriptions for,
• 15:33
  SPEAKER_01: say, thousands, hundreds of thousands of people.
• 15:39
  SPEAKER_01: Get your tasks done by very smart and simple models.
• 15:43
  SPEAKER_01: There's no need for using large models.
• 15:47
  SPEAKER_01: That's a very small, but at the same time efficient way of treating data, treating AI.
• 15:53
  SPEAKER_00: Yeah, but we are Americans.
• 15:56
  SPEAKER_01: We prefer everything in a massive scale.
• 15:59
  SPEAKER_00: If we can flush our toilet with two gallons, we will flush them
• 16:04
  SPEAKER_01: with eight.
• 16:05
  SPEAKER_01: I wasn't like that, but I became an American too.
• 16:08
  SPEAKER_00: If we can drive a small car to work, why not drive it much bigger?
• 16:13
  SPEAKER_00: I'm being critical.
• 16:17
  SPEAKER_00: So the answer, I mean,
• 16:19
  SPEAKER_00: I do think that I understand why people resist, but I definitely agree with you.
• 16:23
  SPEAKER_00: Eventually, it's about sustainability.
• 16:25
  SPEAKER_00: And I think what people understand that there are industries that pollute a lot more.
• 16:31
  SPEAKER_00: I think people think about these industries as so entrenched in everyday life
• 16:38
  SPEAKER_00: that changing them is much harder than blocking something new.
• 16:42
  SPEAKER_00: And as a result, that's what they run to.
• 16:47
  SPEAKER_00: I think there are other ways to think about that, obviously, as a day.
• 16:51
  SPEAKER_00: Has heard about this multiple times because I sometimes get really tired by these arguments.
• 16:56
  SPEAKER_00: And also, again, going back to your previous program points, and that is, let's talk from data.
• 17:02
  SPEAKER_00: Let's not talk from, I get the vibe that this is bad.
• 17:06
  SPEAKER_02: Vaguely complaining about everything and environmental impacts.
• 17:09
  SPEAKER_00: And again, judging it against...
• 17:13
  SPEAKER_00: Other ways of polluting is really instructive.
• 17:17
  SPEAKER_00: How much is it more polluting or less polluting than something else?
• 17:20
  SPEAKER_00: How much is it doing other things?
• 17:23
  SPEAKER_00: Because agriculture, I mean, we talk about how many people are employed in data centers.
• 17:28
  SPEAKER_00: Agriculture employs very few people at the end of the day when we consider how it used to be.
• 17:33
  SPEAKER_00: And so, again,
• 17:34
  SPEAKER_00: it goes back to if it's pasture land,
• 17:36
  SPEAKER_00: how many people would be employed in that land?
• 17:39
  SPEAKER_00: Not very many.
• 17:40
  SPEAKER_00: That's beyond it.
• 17:42
  SPEAKER_00: But I have one last question.
• 17:43
  SPEAKER_00: So you've written about AI and you've talked to us about AI's potential to help,
• 17:48
  SPEAKER_00: especially developing nations, achieve the UN Sustainable Development Goals.
• 17:54
  SPEAKER_00: So in your view,
• 17:55
  SPEAKER_00: when you're thinking about it, what is the biggest barrier for making that happen?
• 18:04
  SPEAKER_01: As I said, AI is helping us with the data, but data is not getting us to the infrastructure.
• 18:12
  SPEAKER_01: In developing nations,
• 18:15
  SPEAKER_01: I think,
• 18:16
  SPEAKER_01: it's again a cliche,
• 18:17
  SPEAKER_01: but the role and the capacity of institutions,
• 18:25
  SPEAKER_01: that concept sits a little bit more concrete than AI itself.
• 18:30
  SPEAKER_01: Because, I mean, we are not expecting AI to build roots or
• 18:41
  SPEAKER_01: Or to bring peace.
• 18:46
  SPEAKER_01: Or all these infrastructure concepts.
• 18:49
  SPEAKER_01: But using the professional data scientists or data engineers, AI engineers,
• 18:59
  SPEAKER_01: In sustainable development projects, we can have valid insights, data-driven insights.
• 19:08
  SPEAKER_01: That's the first part of the project.
• 19:13
  SPEAKER_01: But when it comes to taking actions,
• 19:17
  SPEAKER_01: we need political will and at the same time, the institutional capacity.
• 19:24
  SPEAKER_01: By institutional capacity, I mean the technical proficiency,
• 19:30
  SPEAKER_01: Financial transparency, monitoring, evaluation, stakeholder management, all these fancy words.
• 19:38
  SPEAKER_01: And so when AI is helping us to get the data set done,
• 19:47
  SPEAKER_01: So from the other side,
• 19:48
  SPEAKER_01: if we can combine it with a transparent
• 19:56
  SPEAKER_01: and capable institution.
• 19:58
  SPEAKER_01: So I think we're good to make progress on the development projects
• 20:02
  SPEAKER_01: with our development interventions.
• 20:05
  SPEAKER_00: All right.
• 20:05
  SPEAKER_00: So thank you.
• 20:06
  SPEAKER_01: Thank you so much.
• 20:07
  SPEAKER_02: Thank you so much, Amin.
• 20:09
  SPEAKER_02: Thank you for sharing your thoughts.
• 20:10
  SPEAKER_02: I personally learned a lot.
• 20:12
  SPEAKER_02: Thank you so much.
• 20:13
  SPEAKER_00: So we'll see you next time on Azadeh and Guy
• 20:16
  SPEAKER_02: on AI.