Have we screwed up artificial intelligence?
Just wondering if we're doing justice to AI or we're just leading along because we have to commercialize it as fast as possible. Views are welcome.
8 comments
Big time, and now it's their turn.
Sort of, but not entirely:
- As Papers without Code website presented: many algorithms in papers are manually super-optimized to perform state-of-the-art on benchmarks
- Much of ML work outside of top academia and FAANG lacks rigour or doesn't compare results with "dumb" baselines.
- Many companies (e.g. Google, FB) have not devised a proper recourse procedure for mistakes their "AI" policing bots make (and WILL always make).
However:
- Weren't it for commercialization of ML, we would not have ML accelerators of some sort in almost all modern hardware.
- As Papers without Code website presented: many algorithms in papers are manually super-optimized to perform state-of-the-art on benchmarks
- Much of ML work outside of top academia and FAANG lacks rigour or doesn't compare results with "dumb" baselines.
- Many companies (e.g. Google, FB) have not devised a proper recourse procedure for mistakes their "AI" policing bots make (and WILL always make).
However:
- Weren't it for commercialization of ML, we would not have ML accelerators of some sort in almost all modern hardware.
I'll do you one better. I don't think we've started on AI yet. What we have at the moment is some basic black box decision engines and work data scientists could have done 20 years ago if they'd been given the same budgets.
Artificial neural nets might form one small component of the AIs of the coming years and they will always be useful research tools but they are not the be-all and end-all of AI.
Artificial neural nets might form one small component of the AIs of the coming years and they will always be useful research tools but they are not the be-all and end-all of AI.
> The AI effect occurs when onlookers discount the behavior of an artificial intelligence program by arguing that it is not real intelligence.
> Author Pamela McCorduck writes: "It's part of the history of the field of artificial intelligence that every time somebody figured out how to make a computer do something—play good checkers, solve simple but relatively informal problems—there was a chorus of critics to say, 'that's not thinking'." AIS researcher Rodney Brooks complains: "Every time we figure out a piece of it, it stops being magical; we say, 'Oh, that's just a computation.'"
https://en.wikipedia.org/wiki/AI_effect
> Author Pamela McCorduck writes: "It's part of the history of the field of artificial intelligence that every time somebody figured out how to make a computer do something—play good checkers, solve simple but relatively informal problems—there was a chorus of critics to say, 'that's not thinking'." AIS researcher Rodney Brooks complains: "Every time we figure out a piece of it, it stops being magical; we say, 'Oh, that's just a computation.'"
https://en.wikipedia.org/wiki/AI_effect
No, we haven't screwed it up. What specifically do you think we could have done better?
It's just a problem of managing expectations. The language we use doesn't help either - the term artifical intelligence somehow implies a system that resembles the way humans think. We don't have anything remotely like this. We don't have a faint idea of how to get started on it either.
The machine learning and reinforcement learning techniques that we have are useful in many applications. You shouldn't expect general AI / singularity out of them any time soon and that's fine.
It's just a problem of managing expectations. The language we use doesn't help either - the term artifical intelligence somehow implies a system that resembles the way humans think. We don't have anything remotely like this. We don't have a faint idea of how to get started on it either.
The machine learning and reinforcement learning techniques that we have are useful in many applications. You shouldn't expect general AI / singularity out of them any time soon and that's fine.
Of course we're decades away from the refinement we're looking for, or advertising for that matter. But mainstream AI like Alexa and Google Assistant are selling like hot cakes for the 'human touch.'
Not expecting singularity but for Moore's law to work in a suitable direction, we may have to take a step back and see if commercialization is holding back the development of an even smarter, better, more human AI.
Not expecting singularity but for Moore's law to work in a suitable direction, we may have to take a step back and see if commercialization is holding back the development of an even smarter, better, more human AI.
I believe the language revolving around AI has been construed by those who are not knowledgable technically but spout nonsense to appear knowledgable to attract clicks and sell books.
It doesn't help when laypersons see anatomical terms like neural network and get the image of a computer replicating a human neuron. In fact its merely a fancy for loop. The same results can be achieved with a few nested for loops though not as efficient).
AI at the moment is a term flung around too much. There are some faux AI tools out there to cash in on the trend. Others can be simple implementations of predictive analysis with AI label slapped on the side for grandeur.
Like another poster said, its about managing expectations.
It doesn't help when laypersons see anatomical terms like neural network and get the image of a computer replicating a human neuron. In fact its merely a fancy for loop. The same results can be achieved with a few nested for loops though not as efficient).
AI at the moment is a term flung around too much. There are some faux AI tools out there to cash in on the trend. Others can be simple implementations of predictive analysis with AI label slapped on the side for grandeur.
Like another poster said, its about managing expectations.
We haven’t started yet! Wait some years until the deep learning phase ends and the symbolic AI starts... we’ll be hiring Symbolic Engineers, no more Data Scientists.