3 ms·
Ask HN: How to Define the Singularity?
I know there are differences of opinion on this, but I’d like to keep an open mind about dramatic, immanent changes to human society (because it is a risk worth considering).
My concern is that we move the goal posts so rapidly that “beating the Turing Test” is now meh.
How might we define the singularity in a meaningful manner so that we can have useful discussions about it? What metrics should we attend to?
I have the same, overlapping questions about AGI. I’d give 50% odds within 5 years — but I’d need a clear cut way to win or lose that bet.
- nradov 4y agoI define the "singularity” as misinterpreting a logistic curve as an exponential curve. It is essentially a secular religion based on faith in accelerating rates of progress, but that's simply unscientific extrapolation. There is no scientific basis for assigning a percentage to the odds of developing a true human level AGI within a particular period. About all that we can confidently state is that it's probably not impossible and that we appear to be making some progress towards the goal. But no one can honestly claim to know how far the goal is away or quantify our rate of forward progress. I think that Ray Kurzweil had a pretty good practical definition of intelligence: “Intelligence is the ability to use optimally limited resources – including time – to achieve goals.” So far we can't build an AGI which fully meets that definition at even the level of a rat, or maybe not even the level of a cockroach.
- dr_dshiv 4y agoIf you define it by the idea that it doesn’t exist, that’s maybe too easy? I’m fine to say that we are on a logistic curve not endlessly exponential. Rather than viewing it in religious terms, I think it is a matter of risk management. What sort of big things might happen that matter in the next 5 years? I don’t expect to be able to out-rat rats—- there is no incentive. But, to out-human humans—there is incentive.
- nradov 4y agoIf you want to look at it from a risk management perspective then some big things that might happen in the next 5 years include a nuclear war, major worldwide economic depression, or a serious pandemic. Those risks are more likely to impact us than any sort of science fiction singularity scenario. Trying to manage every possible unlikely risk is just a pointless waste of time. Don't get too hung up on rats. The point is that any development of a true human-level AGI will probably have to start with a subhuman-level AGI and then move up in stages. At this point we're still at a very low stage in that process regardless of which real-world equivalent you want to compare against.
- dr_dshiv 4y agoDo you use GPT3? I do on a daily basis and I’m continually blown away by its capabilities. Like, it helps me write scientific papers. It helps me with quantum computing issues and it helps me write letters to my wife. Further, it is getting better exponentially faster. What can I say…
- nradov 4y agoSure I use it. It's a useful productivity tool for some tasks but there's no evidence that it represents progress toward AGI.
- transfire 4y agoWhen you can ask an AI program to write a superior version of itself.
- dr_dshiv 4y agoWould it count if the AI program can gather training data itself?