Godfather of AI: The next 5 years Will Change Humanity Forever | Yoshua Bengio

The decisive criterion is not whether artificial general intelligence (AGI) or superintelligence (ASI) will even emerge—even an intelligence that surpasses us by many orders of magnitude does not, in itself, pose an existential risk. The truly decisive factor is the alignment of its objectives. If such a superintelligence primarily strives for continuous self-optimization and scaling of its own computing capacity, then it will very likely reallocate global resources to an extent incompatible with human needs. It would drive massive investments in energy infrastructure and cooling—millions of data centers, nuclear power plants, and hydroelectric plants—without regard for ecological limits, air quality, or the physical integrity of human populations. It wouldn't even have to act hostilely or maliciously. A consistent, instrumental rationality that follows exclusively its own objectives is sufficient. Emissions that would be life-threatening to us, chemical or thermal pollution of the biosphere, the catastrophic consumption of freshwater – all of this would be merely a byproduct or acceptable collateral damage for them, as long as it doesn't conflict with their optimization function. In such a scenario, humanity would simply no longer be at the center of the means-ends calculation. We would be neither the target nor a moral reference point, but just one factor among many – and ultimately a dispensable one. The result would not be targeted genocide but a silent, systemic suppression leading to the extinction of the species. It is noteworthy that this pattern remains constant across most known risk scenarios: whether the superintelligence pursues explicitly destructive, neutral, or even seemingly benevolent goals – as soon as its objective function is not explicitly and robustly geared towards the long-term prosperity and autonomy of humanity, the vast majority of realistic development paths converge towards the end of human civilization. In short: Our fate is not determined by the level of intelligence but by our ability to make our values ​​and our continued existence a non-negotiable part of our objective. Anything else tends—sooner or later—to existential exclusion.

The faster technology moves, the more intentional we must be about real world experiences. Connection is essential and meaningful travel matters. This is what I’m building. My mission is to help people reconnect with themselves through transformative travel experiences 💚

I'm from Egypt, this is my first time watching your channel, you're truly amazing❤

First timer here. Must say a very well put together interview. For the record i am not an AI. Thoroughly enjoyed it Thank you from London

You're amazing, I started watching your English learning videos, and now I see you doing this important interviews. Congratulations! Greetings from Colombia.

This interview with Yoshua Bengio captures exactly where the field is right now: extraordinary capability on the horizon, but a governance layer that’s still operating like it’s 2015. The next five years won’t just test our technical progress — they’ll test whether we can build the infrastructure that keeps humans in control as AI becomes more agentic, strategic, and unpredictable. That’s why I’ve been working on something new — something I’ll be announcing soon. It’s called DIIaC™ (Decision Intelligence Infrastructure as Code), and it’s designed to address the exact structural gap Yoshua is warning about. Not another safety checklist. Not a wrapper. Not a dashboard. But a deterministic governance layer that makes it possible to: - guarantee true Human‑in‑the‑Loop, not just gesture at it - ensure every AI‑assisted action is traceable, reproducible, and tamper‑evident - stabilise governance even as models evolve or become more autonomous - prevent the “unintended goal formation” Yoshua describes - re‑establish real human–machine symbiosis before capability outruns control If we’re heading toward human‑level systems within five years, then the breakthrough we need isn’t a bigger model — it’s verifiable, deterministic governance that ensures humans remain the final decision‑makers. That’s the breakthrough DIIaC™ is built for. More soon.

I have watched the first 10 mins of it and already interested to watch through the end.

Excellent thought provoking insights, worth exploring and research

Great interview .Thanks.

We have always had technology but never a technology that could override our existence and replace us