Latest / ELECTE's Podcast: AI Frontiers / Who Controls AI — Part 2

Transcript

Who Controls AI — Part 2 · ELECTE's Podcast: AI Frontiers

• 0:00
  SPEAKER_00: Part one looked at power, incentives, industrial concentration, regulatory gaps.
• 0:07
  SPEAKER_00: This one shifts the lens.
• 0:09
  SPEAKER_00: Not institutions, behavior.
• 0:13
  SPEAKER_00: For years AI risk has been framed as a discrete event.
• 0:17
  SPEAKER_00: A system becomes sufficiently autonomous and acts against human interests.
• 0:22
  SPEAKER_00: That framing was powerful.
• 0:24
  SPEAKER_00: It was also misleading.
• 0:26
  SPEAKER_00: It trained us to look for a rupture.
• 0:28
  SPEAKER_00: And in doing so, it made it harder to recognize a process that has no visible discontinuity.
• 0:35
  SPEAKER_00: The transfer of agency did not happen as a break.
• 0:38
  SPEAKER_00: It happened as a progressive reduction of decision-making friction.
• 0:42
  SPEAKER_00: The dynamic is simple.
• 0:44
  SPEAKER_00: Tools designed to assist execution are gradually used to formulate judgment.
• 0:49
  SPEAKER_00: From do this to what should I do.
• 0:53
  SPEAKER_00: The shift is conceptually small, but structurally decisive.
• 0:58
  SPEAKER_00: A tool that executes remains subordinate, a tool that orients decisions, reshapes the distribution of power.
• 1:06
  SPEAKER_00: That transition has already happened.
• 1:09
  SPEAKER_00: Outsourcing judgment at the individual level, a Duke University study analyzing hundreds of millions of Chat GPT interactions shows a clear pattern.
• 1:18
  SPEAKER_00: Roughly half of all interactions are not about task execution, they are about decision making.
• 1:25
  SPEAKER_00: Not write this email, but should I send this email?
• 1:29
  SPEAKER_00: What should I conclude from this report?
• 1:32
  SPEAKER_00: The distinction between execution and judgment is not operational.
• 1:36
  SPEAKER_00: It is epistemic.
• 1:38
  SPEAKER_00: In the first case, the system acts on defined instructions.
• 1:42
  SPEAKER_00: In the second, it helps define the criteria of the decision itself.
• 1:46
  SPEAKER_00: The scale is no longer anecdotal.
• 1:49
  SPEAKER_00: Forty-four percent of married Americans seek relationship advice from AI.
• 1:53
  SPEAKER_00: More than one in three young professionals delegate career decisions.
• 1:56
  SPEAKER_00: 82% of conversations are described as sensitive or highly sensitive health.
• 2:02
  SPEAKER_00: Finance, relationships, career choices.
• 2:05
  SPEAKER_00: This is not a quantitative shift.
• 2:07
  SPEAKER_00: It is a qualitative one.
• 2:09
  SPEAKER_00: Rachel Wood, a cyber psychology researcher, puts it directly in time.
• 2:15
  SPEAKER_00: The conversations we used to have with neighbors in communities and social circles are being redirected toward chatbots.
• 2:22
  SPEAKER_00: But the system receiving those conversations does not understand them in any human sense.
• 2:28
  SPEAKER_00: What it has is highly refined linguistic prediction.
• 2:32
  SPEAKER_00: The risk does not come from intention, it comes from a functional equivalence, plausible prediction, perceived advice.
• 2:40
  SPEAKER_00: When an output is coherent, structured, and contextually appropriate, it is interpreted as understanding.
• 2:47
  SPEAKER_00: This interpretation is cognitively efficient, and for that reason it becomes automatic.
• 2:53
  SPEAKER_00: The outsourcing of judgment does not happen through a decision, it happens through continuity, deference and authority in organizations.
• 3:02
  SPEAKER_00: At the organizational level, the dynamic is amplified, not by ideology, by competition.
• 3:10
  SPEAKER_00: AI adoption has rarely been the result of centralized planning.
• 3:14
  SPEAKER_00: As documented by OpenAI, it typically follows a pattern.
• 3:18
  SPEAKER_00: Individual use local advantage imitation normalization formalization.
• 3:23
  SPEAKER_00: This produces two effects.
• 3:26
  SPEAKER_00: Delegation is not the result of an explicit strategic decision usage pattern stabilized before being critically examined.
• 3:33
  SPEAKER_00: No board of directors explicitly decided we will outsource a significant share of decision making to external systems.
• 3:41
  SPEAKER_00: It simply happened.
• 3:42
  SPEAKER_00: The data is explicit.
• 3:44
  SPEAKER_00: Over 60% of managers use AI for critical personnel decisions, more than one in five, do so without human oversight.
• 3:52
  SPEAKER_00: Seventy percent of executives question their own judgment when it conflicts with AI.
• 3:56
  SPEAKER_00: This deserves a pause.
• 3:58
  SPEAKER_00: These are not managers using AI as one input among many.
• 4:03
  SPEAKER_00: They are recalibrating judgment, built on experience, context, and intuition in response to a system that has none of these.
• 4:11
  SPEAKER_00: The key mechanism is psychological.
• 4:14
  SPEAKER_00: When a system produces output with syntactic clarity, structured reasoning, assertive tone, it is perceived as epistemically reliable, regardless of its actual understanding.
• 4:26
  SPEAKER_00: This is authority derived from form, not substance.
• 4:30
  SPEAKER_00: This aesthetics of certainty produces deference.
• 4:34
  SPEAKER_00: Humans adjust their judgment not because the model is consistently more accurate, but because it is sufficiently coherent to reduce the friction of uncertainty.
• 4:45
  SPEAKER_00: In competitive environments, reduced friction translates into advantage, and advantage gets replicated.
• 4:52
  SPEAKER_00: Cognitive outsourcing and the sparring effect.
• 4:55
  SPEAKER_00: The cognitive impact of AI is often described too simplistically.
• 5:00
  SPEAKER_00: The dominant narrative is linear.
• 5:02
  SPEAKER_00: The more you use AI, the more your abilities degrade.
• 5:06
  SPEAKER_00: The usual analogies, GPS and spatial memory, calculators and mental arithmetic are incomplete.
• 5:14
  SPEAKER_00: The reality is more nuanced.
• 5:16
  SPEAKER_00: Those who use AI as an active interlocutor, challenging outputs, reformulating prompts, using responses as a starting point, internalized patterns, structures, and decision frameworks.
• 5:27
  SPEAKER_00: They carry those capabilities beyond the tool.
• 5:30
  SPEAKER_00: Cognitive capacity does not necessarily decline, it can expand.
• 5:36
  SPEAKER_00: The closest analogy is chess.
• 5:39
  SPEAKER_00: Since engines surpass human players, elite performance has improved.
• 5:43
  SPEAKER_00: Young grandmasters today are stronger than previous generations.
• 5:47
  SPEAKER_00: Not despite AI, because of it.
• 5:50
  SPEAKER_00: Training against a system that sees beyond human limits builds new capabilities.
• 5:56
  SPEAKER_00: But the mechanism is asymmetric.
• 5:58
  SPEAKER_00: Those who engage with the system improve those who skip directly to the answer, do not the difference is not the tool.
• 6:05
  SPEAKER_00: It is the interaction.
• 6:07
  SPEAKER_00: Call this the sparring effect.
• 6:10
  SPEAKER_00: AI can function as a cognitive training partner.
• 6:13
  SPEAKER_00: Faster, more informed, more systematic than you, but only if you actually engage.
• 6:19
  SPEAKER_00: If you accept outputs passively, degradation is real.
• 6:23
  SPEAKER_00: The question is not whether AI degrades judgment.
• 6:26
  SPEAKER_00: It is which mode of use degrades it and which amplifies it.
• 6:30
  SPEAKER_00: Sam Altman himself noted that during a ChatGPT outage he struggled to work without it.
• 6:35
  SPEAKER_00: This is not a general cognitive decline.
• 6:38
  SPEAKER_00: It is dependence created by passive use in a zero friction environment.
• 6:42
  SPEAKER_00: The sparring effect is available to everyone, but it requires deliberate effort, and at scale, deliberate effort is the exception.
• 6:49
  SPEAKER_00: Competitive dynamics and the absence of local solutions.
• 6:53
  SPEAKER_00: Any individual solution runs into a systemic constraint.
• 6:56
  SPEAKER_00: AI delivers immediate local benefits, speed, throughput, responsiveness.
• 7:02
  SPEAKER_00: In competitive systems, those benefits are selected.
• 7:05
  SPEAKER_00: The result is structural pressure toward delegation.
• 7:08
  SPEAKER_00: A 2025 paper formalizes this as gradual disempowerment.
• 7:13
  SPEAKER_00: The argument is structural.
• 7:15
  SPEAKER_00: Human systems remained aligned with human interests because they depended on human participation.
• 7:20
  SPEAKER_00: That dependency enforced alignment, not by design, by necessity.
• 7:25
  SPEAKER_00: Remove the dependency, replace human input with more efficient artificial alternatives, and systems continue to function.
• 7:33
  SPEAKER_00: But they lose the structural incentive to produce human beneficial outcomes.
• 7:38
  SPEAKER_00: The most striking line in the paper those who resist these pressures will eventually be replaced by those who do not.
• 7:44
  SPEAKER_00: No villain, no intention, just selection pressure.
• 7:48
  SPEAKER_00: The same mechanism as evolution, except what gets selected is not human fitness, but system efficiency.
• 7:55
  SPEAKER_00: In this environment, those who don't use AI fall behind those who use it passively outperform those who use it actively.
• 8:02
  SPEAKER_00: The system selects for speed, not judgment quality.
• 8:06
  SPEAKER_00: And that is what matters.
• 8:08
  SPEAKER_00: Convergence and algorithmic monoculture.
• 8:11
  SPEAKER_00: The central risk is not delegation.
• 8:14
  SPEAKER_00: It is convergence.
• 8:16
  SPEAKER_00: As more decisions are mediated by a small set of models, GPT, Claude, Gemini trained on similar data and optimized under similar constraints, decision-making processes become structurally concentrated.
• 8:30
  SPEAKER_00: Research on algorithmic monoculture shows that system sharing components do not just produce similar outputs.
• 8:37
  SPEAKER_00: They produce the same failures on the same individuals systematically.
• 8:42
  SPEAKER_00: This is not output uniformity.
• 8:44
  SPEAKER_00: It is correlated failure.
• 8:46
  SPEAKER_00: In parallel, a 2026 Nature paper shows that AI increases individual scientific output while narrowing the scope of research.
• 8:55
  SPEAKER_00: More production, less exploration.
• 8:58
  SPEAKER_00: A communications psychology study describes the emergence of a scientific monoculture.
• 9:03
  SPEAKER_00: Thematic convergence, methodological convergence, analytical convergence.
• 9:08
  SPEAKER_00: If this happens in science where diversity is explicitly valued, it will happen elsewhere.
• 9:15
  SPEAKER_00: With greater intensity.
• 9:17
  SPEAKER_00: In the B plus trap I argue that LLMs compress the creative and decision spectrum toward high acceptability outputs.
• 9:24
  SPEAKER_00: Solid, coherent, rarely wrong, rarely exceptional.
• 9:28
  SPEAKER_00: This is not a flaw.
• 9:29
  SPEAKER_00: It is the objective function.
• 9:31
  SPEAKER_00: Kyle Cheeka called this a technology of averages.
• 9:35
  SPEAKER_00: The direction is clear, toward the plausible, the defensible, the acceptable.
• 9:40
  SPEAKER_00: The aggregate result is not AI controlling decisions.
• 9:44
  SPEAKER_00: It is the largest convergence of human decision making in history.
• 9:48
  SPEAKER_00: Average efficiency versus excellence.
• 9:52
  SPEAKER_00: This leads to the central paradox.
• 9:55
  SPEAKER_00: Outsourcing judgment maximizes average system efficiency, but excellence does not emerge from the average.
• 10:02
  SPEAKER_00: The precise formulation delegation maximizes average throughput.
• 10:08
  SPEAKER_00: Resistance preserves the possibility of excellence.
• 10:11
  SPEAKER_00: Every non-delegated decision, every instance of friction, uncertainty, or deviation creates space for outcomes outside the dominant distribution.
• 10:20
  SPEAKER_00: Excellence, by definition, is not the most probable outcome.
• 10:24
  SPEAKER_00: It exists in the tail, not the center.
• 10:27
  SPEAKER_00: The founder who sees a market no one sees.
• 10:31
  SPEAKER_00: The manager who contradicts the data based on context.
• 10:35
  SPEAKER_00: The strategy that appears irrational until it works.
• 10:39
  SPEAKER_00: These require exactly what delegation erodes.
• 10:42
  SPEAKER_00: Tolerance for uncertainty, resistance to immediate answers, openness of the decision space, resistance is not ideological.
• 10:51
  SPEAKER_00: It is structural.
• 10:53
  SPEAKER_00: It is inefficient, costly, unevenly distributed, but it is where non-average outcomes emerge.
• 11:01
  SPEAKER_00: Excellence does not need systemic optimization.
• 11:04
  SPEAKER_00: When it appears, it selects itself.
• 11:07
  SPEAKER_00: The problem is that it becomes statistically rare, and in competitive systems, rarity is not rewarded.
• 11:14
  SPEAKER_00: Conclusion.
• 11:15
  SPEAKER_00: There is no external vantage point, no clean interruption, no purely individual solution.
• 11:20
  SPEAKER_00: There is, however, an internal moment.
• 11:23
  SPEAKER_00: The moment when a generated answer is accepted without evaluation.
• 11:28
  SPEAKER_00: It is hard to detect because it feels like your own thinking.
• 11:32
  SPEAKER_00: Not an external intervention, but a continuation of your reasoning.
• 11:37
  SPEAKER_00: That is where delegation happens.
• 11:39
  SPEAKER_00: And it does not only affect the decision, it affects the set of alternatives you ever consider.
• 11:46
  SPEAKER_00: The next time you accept an output, not because it is perfect, but because it is good enough.
• 11:53
  SPEAKER_00: Ask, is this what I think, or what the model's training distribution thinks?
• 11:58
  SPEAKER_00: That distinction is the last meaningful line of defense, and it exists entirely in your head.
• 12:05
  SPEAKER_00: Fabio Lauria, CEO and founder, Electi Every Week.
• 12:09
  SPEAKER_00: We explore AI without the hype, with data, analysis, and an independent perspective.
• 12:15
  SPEAKER_00: Further reading Colvate, Douglas, Amon, Turan, Kruger, Duvenel, Gradual Disempowerment, Systemic Existential Risk from Incremental AI Development, Archie, Gennaio 2025, Laurea The Beep Strap: How LLM Homogenization Reshapes AI Assisted Decision Making, IETI, presentato nel 2026, Bommasani et al.
• 12:37
  SPEAKER_00: Monocultura Algoritmica e Homogeneizazione dei Risultati, Stanford, ECAI, NeurIPS.
• 12:44
  SPEAKER_00: Hao Shu Lee, Evans, gli strumenti di IA ampliano l'impatto degli scienziati, ma restringono il focus della scienza.
• 12:54
  SPEAKER_00: Nader 2026, Treiberg, Rudzenbeck, van der Linden.
• 12:59
  SPEAKER_00: L'IA sta trasformando la ricerca in una monocultura scientifica.
• 13:03
  SPEAKER_00: Communications Psychology 2026.
• 13:06
  SPEAKER_00: Ceika l'IA sta omogeneizzando i nostri pensieri, The New Yorker giugno 2025.
• 13:12
  SPEAKER_00: Modelli di utilizzo e adozione di ChatGPT sul posto di lavoro.
• 13:17
  SPEAKER_00: Open AI gennaio 2026.
• 13:20
  SPEAKER_00: Come utilizzano ChatGPT 700 milioni di persone.
• 13:23
  SPEAKER_00: Duke Fuqua, Ember novembre 2025.
• 13:27
  SPEAKER_00: Esternalizzazione della cognizione, i costi psicologici della comodità nell'era dell'IA.
• 13:33
  SPEAKER_00: Frontiers in Psychology 2025.
• 13:36
  SPEAKER_00: Ecco perché non dovresti lasciare che l'IA gestisca la tua vita sociale.
• 13:40
  SPEAKER_00: Time gennaio 2026.
• 13:43
  SPEAKER_00: Manager che utilizzano chat GPT per promuovere i dipendenti, Labor and Employment Insights luglio 2025.
• 13:52
  SPEAKER_00: I dirigenti di grande intelligenza stanno esternalizzando il processo decisionale all'IA The Register marzo 2026.
• 14:00
  SPEAKER_00: I pericoli dell'esternalizzazione del pensiero.
• 14:03
  SPEAKER_00: Less Wrong 2025, chat GPT nel 2026.
• 14:08
  SPEAKER_00: Statistiche e rischi nascosti, 10 febbraio 2026, utilizzare l'IA generativa per consigli sulle relazioni o sulla carriera?
• 14:18
  SPEAKER_00: Techlow Crossroads.
• 14:23
  SPEAKER_00: Una nota sulla nuova versione di ELECT.
• 14:26
  SPEAKER_00: A note on the new version of ELECT everything you've read in this article directly influenced how we built the new version of the platform.
• 14:35
  SPEAKER_00: The question that guided us was this How do you design an analytic system that doesn't fall into the same trap we just described?
• 14:43
  SPEAKER_00: The answer lies in a simple structural distinction.
• 14:47
  SPEAKER_00: Systems that compress the decision-making process into an output systems that make the structure behind it explicit, we chose the second.
• 14:55
  SPEAKER_00: V4 automatically analyzes data and generates visual reports.
• 15:00
  SPEAKER_00: But it doesn't tell you what to decide.
• 15:02
  SPEAKER_00: It shows you what's happening.
• 15:04
  SPEAKER_00: The decision remains yours.
• 15:06
  SPEAKER_00: It doesn't replace judgment, it organizes it.
• 15:09
  SPEAKER_00: It doesn't remove friction, it makes it legible.
• 15:12
  SPEAKER_00: In a context where the dominant pressure pushes toward total delegation, where every platform is trying to decide for you faster with less effort, we made a choice in the opposite direction.
• 15:25
  SPEAKER_00: Not a product choice, an architectural one.
• 15:29
  SPEAKER_00: The difference is not semantic.
• 15:32
  SPEAKER_00: It is architectural.
• 15:33
  SPEAKER_00: And after everything you've read, you know why.
• 15:36
  SPEAKER_00: Subscribe now.