‹ BackHN Continuity

Thread

It's Time to Investigate the AI Labs

629 points · 277 comments · ibobev

  1. jimmyjazz14 · · focus · HN ↗
    > We must move past vague discussions of “AI” and instead isolate the specific types of systems that are creating problems.

    This is absolutely correct and its surprising how rarely you see commenters in the media push to go into the specifics on this. AI is just matrix math and its what you connect that math to that matters, we should talk a little more about what we are willing to connect AI to and little less about how scary or capable the models appear.

    1. yodsanklai · · focus · HN ↗
      A point that Yann LeCun has been making for a while is that it's meaningless to regulate AI, but we want to regulate applications. For instance, you don't want autonomous vehicles to hit people, whether they use AI or not. In practice, most applications of AI are already regulated.
      1. thaumasiotes · · focus · HN ↗
        > For instance, you don't want autonomous vehicles to hit people, whether they use AI or not.

        What would it mean for a vehicle to be "autonomous", but to "not use AI"?

        1. ehe78qhe · · focus · HN ↗
          Trains are often highly automated without AI, for example. Aircraft are often automated for most of their flight without AI.
          1. throwawayffffas · · focus · HN ↗
            Autopilots perceive the environment, reason, plan and execute actions. Autopilots are AI.
            1. ehe78qhe · · focus · HN ↗
              Most autopilots don't involve reasoning and planning. In fact, autopilots predate digital computers. The first ones were developed prior to WW1.

              If a basic heading hold PID loop is AI so is a thermostat from 1990. At which point the word is so meaningless it is obviously a useless definition.

              1. throwawayffffas · · focus · HN ↗
                Ofcourse they do reasoning and planning, it's entirely encoded in software ofcourse, but modern autopilots don't just hold a course, they perform maneuvers and follow waypoints the process of turning the objective turn 35 degrees to actuator inputs is reasoning and planning.

                AI is not magic it's software, the definition is far from pointless, google maps giving you a route to your destination is AI.

                1. dminik · · focus · HN ↗
                  My recipe book has 300 AIs and all are delicious. Google Maps running A* or whatever is obviously not AI. Come on.
            2. wongarsu · · focus · HN ↗
              In the academic definition of AI (which includes things like pathfinding with A* or Djikstra's algorithm) autopilots are definitely AI.

              The popular definition of AI seems to shift every couple of years to follow hype cycles. Chess computers used to be AI (and fit your definition), now they are not. Not sure if Resnet is still AI today, or if autopilots still qualify

            3. gizajob · · focus · HN ↗
              You can’t say to an autopilot “create an excel spreadsheet of the current vectors and conditions and upload it to google docs”.
              1. throwawayffffas · · focus · HN ↗
                You can't say that to an mri machine that detects and highlights tumors. You can't say it to jev either, nor googles spam filters. But all these are built in the same way as any LLM. The term AI is not restricted to chat based LLMs.
                1. gizajob · · focus · HN ↗
                  None of that applies to autopilots though and it’s disingenuous for anyone to say that an autopilot is an AI rather than an automated system designed for a very specific task.

                  An autopilot is cybernetic without being intelligent.

                  1. throwawayffffas · · focus · HN ↗
                    Read wongsarus comment above.
              2. Sharlin · · focus · HN ↗
                That is neither a sufficient nor necessary condition for something to be AI.
                1. gizajob · · focus · HN ↗
                  I was giving a framing around its complete lack of generality or intelligence. Rather than contemporary definitions of AI shifting, what is being shown is that previous definitions of AI were misapplied in the hope that the field might bootstrap itself into existence, which it’s done.

                  As I’ve said elsewhere - an autopilot is cybernetic, it responds to inputs and stimulus in data but it doesn’t possess anything we’d now regard as falling under intelligence because it’s a strictly coded deterministic system and designed for it’s specific environment. I couldn’t take the autopilot from a Boeing and put it into an Airbus and hope for any kind of useful result. I could take a human pilot trained on a Boeing and put them in an Airbus or a Cessna, and if push came to shove in an emergency, they could probably do a decent job of getting the thing pointed in the right direction and landed on the ground.

            4. toofy · · focus · HN ↗
              i think you might accidentally be conflating an algorithm with the colloquial use of ai.
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.