AI and Existential Risk

Tl;Dr: What’s different about _this_ existential risk? Inerrant risk of tech

There’s been a lot of talk about AI Risks – things like they may take jobs, the economy may crash, bad people might use them to do bad things, or the AI Agents might do bad things themselves.

And if there are risks, then we might decide to apply governance (a topic for a different day).

There are a LOT of potential AI Risks. The MIT AI Risk Repository break them down into 7 categories:

1. Discrimination & toxicity – AI Agents discriminating against groups of people (e.g. sexism), or spreading toxic information (AI harassing people)
2. Privacy & Security – AI agents finding and leaking private information
3. Misinformation – AI spreading incorrect information (e.g. earth really IS flat)
4.  Malicious actors & misuse – enabling bad people to do bad things (faster/bigger)
5. Human-computer interaction – people becoming too dependent on AI (‘I can’t do basic calculations any more, let’s as AI)
6. Socioeconomic & environmental harms
7. AI system safety, failures & limitations

I’d like to focus on existential risks – the ones that get the most media time and are the scariest. The ones that would cause humanity (or a whole civilization or society) to collapse just die out.

Existential risks are scary because they are so final. You can imagine recovering from limited risk. But when Chicken Little says the sky is falling, presumably there is no coming back from that.

An important question is: Have we (humans, societies) ever faced existential risks before? If so, what happened? And if not, what makes these risks different than ones we’ve experienced before.

The biggest existential risk to homo sapiens happened between 800,000 and 900,000 years ago. Our species was reduced to less than 1500 individuals for a period of 120,000 years. The speculative cause was environmental changes in Africa.

Wikipedia cites two other bottlenecks, even more recent: only 75,000 years ago, it seems as if the total number of individuals was between 10 and 30k.

Obviously sapiens rebounded from these incidents. The questions before us today are:

• How high is the risk of existential crisis and/or population bottleneck
• How many potential causes are there
• How many have we survived in recent history and…
• How have we survived them?

The overall question is: How could we navigate such a risk to avoid massive loss of life or extinction.

And… you don’t have to turn the clock back all that far to find examples.

7 minutes to midnight

In 1947 the Bulletin of Atomic Scientists designed the Doomsday Clock to warns the public about how close we are to destroying our world with dangerous technologies of our own making. Here you can see clock values from 1947 (7 minute) until 2026 (1 minute, 15 seconds).

Obviously, if the clock ever reached zero there would be nobody to update it. And it’s notable that the timeline has gotten shorter since it started – we’re now at roughly 20% of the 7 minutes given back in 1947.

Annnd… in 1991, the value was at it’s max height – 17 minutes.

At that time, the primary existential risk was Nuclear Annihilation. Over the course of 50-some-odd years, humanity managed to more than double the amount of time on the clock through concerned global governance of Nuclear weapons.

Is this a useful analogy? Could we apply this kind of thinking to current AI risks? How are things similar, how are they different and what can we learn?

No answers today, maybe in the next post…