Cheryl Rofer, a retired nuclear chemist who is currently acting as servant to a pair of adorable cats, wants to know why I think artificial intelligence is so great:
I keep asking that question and getting no answers. There’s a lot of activity going on in what is called artificial intelligence, but it is more accurately called machine learning. The chatbots now amazing the credulous simply predict, on the basis of a training set, which word will follow the previous one.
....But there has to be a reason that Big Tech is putting money into the enterprise, and presumably they believe they will make money out of it. They gloss this with Benefits To The User. Better search! Automated letter writing! They leave out things like automated facial recognition to help arrest people and yet another way to separate the customer from the seller.
This is a tricky question. I happen to think that things like ChatGPT already have some usefulness, including—yes—better search and help with letter writing, among other things. But I also agree that the current version of ChatGPT isn't anywhere close to artificial intelligence.
But! Even though I think true AI is still ten or fifteen years in the future, the software we build along the way will become increasingly useful even if it isn't truly our cognitive equal. I've suggested, for example, that in a couple of years ChatGPT might very well be good enough to replace lots of lawyers. Probably some doctors, too. Maybe even teachers, though that's more speculative.
Those are just tidbits, though. The real usefulness of current machine learning software is that it's a (necessary) step along the way to AGI—artificial general intelligence. AGI will be able to do pretty much anything humans can do, and shortly after that it will be able to do more than humans can do.
Now, if you don't believe that we'll ever be able to produce AGI, or that it's hundreds of years away, then there's nothing to talk about. We'll just have to wait a decade or two and see what happens.
But if you agree with me that we'll have it fairly soon, then Cheryl's question becomes: What do I think is the usefulness of a machine version of human intelligence? And there, I assume the answer is fairly obvious:
- Since I'm positing that AGI-powered robots will be the cognitive equal of humans, then by definition they'll be able to perform every conceivable human job.
- But better than humans! In addition to our raw smarts, they'll work 24 hours a day, have instant access to all human knowledge, and never ask for a raise.
- As teachers, they'll be personalized to every student. It won't matter if you have a visual learning style. It won't matter if you're on the spectrum. It won't matter if you're dyslexic or ADHD. They'll adapt to whatever works best.
- As caregivers they'll be completely reliable, infinitely patient, and willing to talk endlessly about anything that we olds want to natter on about.
- If we haven't done so already, superhuman AGI will solve both our energy problems and global warming.
I'll confess that I've never understood why anyone would ask what AGI is good for. It just seems so obviously world-changing. But maybe I'm missing something.
There's an assumption that AGI can be used to figure out better ways of doing things that we already do. For example, it could design a better building, improve a chemical process or do a better job of translating poetry. Right now, the approach is to write software that understands buildings, chemical processes or poetry in different languages. This requires people who understand the subject matter and know how to write software.
A useful AGI has to be able to learn that subject matter by itself and then write or emulate software for solving the requisite problem. Figuring out the next word to produce an almost coherent sequence of words is only a small part of the problem. ChatGPT strikes me as a distraction. A textual answer would be nice, particularly for something like translating poetry, but if an AI could provide good answers in a stylized domain appropriate manner, like an architectural drawing or a process diagram.
I keep waiting for someone to address the real problem. A useful AI system has to be able to instantiate a world model and reason about it. An AGI has to be able to build the model itself and determine the appropriate modes of reasoning. Even a brief conversation with ChatGPT reveals that it is incapable of creating and dealing with a world model. It's not going to design a very good building if it can't keep track of where the structural supports are, and it isn't going to translate poetry well if it can't reason about the way it works at multiple levels.
I won't say ChatGPT is a complete waste of time and talent. Being able to scan a large corpus and exploit parallel semantic structures is an interesting demonstration, but it isn't clear what it might be useful for let alone how it could be extended to actually do something useful. I see no problem with your AGI desiderata, but I see now way that ChatGPT gets us more than a micron closer to the nearest star.
Google paid 99 dollars an hour on the internet. Everything I did was basic Οnline w0rk from comfort at hΟme for 5-7 hours per day that I g0t from this office I f0und over the web and they paid me 100 dollars each hour. For more details
visit this article... https://createmaxwealth.blogspot.com
I doubt we'd be able to shift culturally fast enough to avoid widespread poverty and misery during the conversion to Drum's AI utopia.
But even if we reached it, humans would probably just peter out. Who would want to have actual babies, when you could have a robot AI child who never has a temper tantrum and will happily sit on your lap while you read just as many books as you want to (and not the same one for the 300th time)?
To think AI will benefit humanity is to assume its owners want it to. Me, I think they will use it to enrich themselves further. Lay off workers. Sow disinformation. Identify dissidents. And so on.
And beware of the thought that humans will always be needed as consumers. Oligarch A will not give people the money to buy from oligarch B and vice versa.
Finally, the state cannot redistribute what is not there. Production already shifts too easily to low-wage and low-tax countries. Imagine the race to the bottom that will ensue when you can escape redistributive taxes simply by moving machines, or even by uploading software to a tax haven.
You're assuming the continuation of existing economic structures under circumstances where those structures are obviously inapplicable. Wide spread human level AI means the end of scarcity, which means the end of all existing economic systems because they are all ways of distributing scarce resources. Getting there from here will undoubtedly involve some turmoil; in the worst cases, it will involve some guillotines or equivalent.
As far as the race to the bottom is concerned, sooner or later you get there. When it becomes to produce goods and services without workers, goods and services will be produced everywhere; there won't be poor places and rich places.
The end of scarcity? I thought we already reached that, except we found ways to keep people hungry by wasting food and other ways to make people feel scarcity over things they don't really need. Heck, scarcity is what makes money and for that reason it ain't going away. You appear to assume that the winds of time will blow the deeds of decision-makers away. But AI is a deed of decision-makers and it is done in their interest.
Yes, I assume the continuation of capitalism, and expect it to get worse, as ever more power shifts from labor to capital.
The thing that concerns me with all AI development is that the AI is likely to be tuned to benefit the people who paid for it, not the people who used it.
Yes, you can have a robot nanny that takes care of your kid, but is constantly nagging you and trying to sell you stuff that will "make your child's life better". Just for instance. I mean, The Algorithm of Facebook (and correspondingly with YouTube) is now understood to be tuned to increase engagement, which is not at all the same as pleasure or satisfaction. Someone who is satisfied doesn't look at more clicks or videos.
Like ruralhobo, you are assuming the continuation of capitalism. That is extremely unlikely.
Kevin said:
"What do I think is the usefulness of a machine version of human intelligence?"
Why would we assume it will be a machine version of human intelligence. This is an anthropomorphic formulation of the problem.
Would not a general version of machine intelligence, even if it was not like human intlligence, be useful?
Indeed why do we assume that human intelligence is so special and assume that should be the desired end goal. Indeed human intelligence is fairly poor quality intelligence if we look at it objectively: it is clouded by emotion, it jumps to unsupported conclusions, and has a self-confidence in its correctness which is often/usually not justified.
I think I will be happy with a general machine intelligence: it does not have to be similar to human intelligence.
I made that comment in another of Kevin's columns on AI. There is a science fiction definition of "intelligent alien" as " a being that thinks as well as a human but not like a human". The odds are quite high that developments in AI will produce artificial aliens, not artificial humans.
( {sigh} This, again...?)
Go look at some software. Any software.
I'll wait.
Take a look at a blue-green algae and try to envision a human from it. It took two billion years to get here from there; machine evolution proceeds faster.
What we've seen in personal computers is an example. I can remember when 64k was a lot of memory for a PC; now, about forty years later, one million times that is not unheard of. There used to be a CPU performance benchmark called Norton SI; a 4.77 mhz Intel 8088 CPU had a performance rating of one. The hottest CPUs were up to about 100 when that benchmark was abandoned; if it could be meaningfully applied to modern CPUs, which it probably can't, you would see SI ratings in tens of thousands, maybe hundreds of thousands. All of that has happened in forty years. What will the next forty years bring?
Sabine Hossenfelder believes chatbots exhibit a level of inherent understanding,
https://www.youtube.com/watch?v=cP5zGh2fui0
You forgot one.
https://tvtropes.org/pmwiki/pmwiki.php/Main/TheRuleOfFirstAdopters
Pingback: The “AI” Challenge – My Response – Lawyers, Guns & Money
But better than humans! In addition to our raw smarts, they'll work 24 hours a day, have instant access to all human knowledge, and never ask for a raise.
Why would you think this? Why would a human-equivalent AGI be able to run 100% uptime indefinitely? Why would it necessarily have access to all human knowledge? Why, if it had even as much smarts as a human, would it not ask "what's in this for me?"
In answer to your last question, we're not assuming artificial emotion. "What's in it for me" is a question based in emotion, not reason.