On the news today, they talked about a new fighter jet — the F-47.
The person they interviewed made a comment that sounded like what AI was supposed to be. He said that when the pilot does something, the jet (system) learns from it. That sounds like real AI to me. Going beyond what it's taught. Actually learning from experience.
The person they interviewed made a comment that sounded like what AI was supposed to be.
A lot depends on what you think AI is supposed to be.
Is your view of what AI is supposed to be based in real world computer science.
Or is you view of what AI is supposed to be based in science fiction, sentient computer systems and androids.
If it's the latter, nothing we have today comes anywhere close.
The biggest super computers that exist today are orders of magnitude short of the processing capacity of a human brain.
Computers seem much more powerful because:
1. They are blindingly fast.
2. They can focus on a single task/computation.
Most people drastically underestimate the volume of sensory data our brains have to constantly process and the complexity of tasks we perform every day without actively thinking about them.
We need a clear, objective definition of what AI is supposed to be before we can judge the "AI" in the news piece you are referring to.
A lot depends on what you think AI is supposed to be.
What the computer scientists said is was going to be when it was first announced.
The example they used at the time was playing chess. They claimed the AI program could be taught the moves and basic strategy. Then the AI computer would play game after game, learning from each, until it could beat anyone who taught it how to play. It would actually learn.
That was the promise of AI that never materialized and is now only SciFi. But if the AI in the F-47 jet can learn, well…. Then again, are they using the term "learn" misleadingly?
More likely, like most modern AI software, the code is SO buggy, they're hoping it can somehow 'self-correct' it's own hallucinations, yet recent experience shows, that's highly unlikely.
What we need, is more of the old-fashioned, working AI, and less of the 'hallucinatory' pie-in-sky hyperbole.
Computers are faster at doing multiple small tasks, yet trying for anything but that is mostly pipe dreams.
Computers are faster at doing multiple small tasks, yet trying for anything but that is mostly pipe dreams.
Not necessarily. I remember being involved in the early days of data warehouse. We, the designers, stored massive amounts of data in databases in relationships. The data warehouse software analyzed that data and relationships and came up with stuff no one thought of.
I remember one example that surprised a lot of people. It recommended that a grocery store put six-packs of beer next to disposable diapers. Everyone thought that was nuts until someone figured out why the software was saying to do that. It seemed a wife with a baby would call her husband and tell him to pick up a box of Pampers on the way home from work (yeah, different times). When he went to get the diapers, he saw the beer and bought that too.
AI was the next thing. The data warehouse could only analyze existing data and relationships. It couldn't "think" and learn. The promise back then for AI was to think and learn. That never happened.
Isn't that an expert system? It memorises what pilots do under what circumstances. Given new circumstances that no pilots have experienced during its time, the system won't know what to do. Shades of the Tesla car that crashed because its censors detected plain white ahead.
AJ
Technically, it drove faster than its sensors were designed to process. Yet it still sold like hotcakes, until public sentiment turned on its 'inventor' from profiting while everyone was losing jobs and its sales virtually dropped overnight.
Fate, and the public, can be cruel at times, fickle beasts that we are.
Technically, it drove faster than its sensors were designed to process.
The Tesla engineers reported that the truck was spotted 10.5 seconds before impact but it wasn't correctly identified, and the car could have have stopped in time up to 1.5 seconds away.
AJ
The only ones losing jobs were government bureaucrats, who were being hired like hotcakes when the rest of us struggled.
At the risk of 'current politics', federal government hiring has been very low compared to private sector hiring over the last decade or so. Aside from temporary seasonal hiring in census years, which creates odd spikes, new government-bureaucrat hiring at the federal level in the United States over the past decade is roughly one-fifteenth of private-sector hiring. Over the last five years, it's under one-twentieth (more federal jobs added in that five years than the previous five, but far more private sector jobs added in that five than in the previous five, ignoring temporary 2020-related job losses and refills). That doesn't feel much like 'hotcakes' to me.
State job growth is higher, but the comment was relative to federal jobs, since those are the ones being lost.
Most of the ones losing their jobs aren't the high level bureaucrats, they are the ones that were doing the work. I know of several medical people from an Arizona VA medical center that were "surplused." They let enough go that the wait time for an appointment has about doubled. If they fire doctors and nurses, who will care for our Vets?
In the interest of eliminating DEI, they closed the Native American meditation rooms, but kept the Christian & Jewish chapels and chaplains. Go figure! How much money are they going to save by locking a couple of doors.
That sounds like real AI to me. Going beyond what it's taught. Actually learning from experience.
I'd call that adaptive machine learning — it's extrapolating from its inputs based on an algorithm, and within a set of defined parameters which may not be exceeded.
It is not real AI. They are programs that can learn and adapt. but it's still following it's algorithm. It's just taking in new data when making it's decisions based on its programming.
Computers can beat any human at chess. They don't think about the game like humans do, but they've been programmed with such huge volumes of data that no human can keep up any more. Do the modern computer masters remember anything about their prior experiences with any given player? Probably not. In that respect, they probably don't "learn". But they did "learn" by ingesting all of that data in the first place.
Modern LLMs like ChatGPT don't, as I understand it, learn in real time. They learn by periodically undergoing training, which is typically separate from regular user interactions. But I expect they could, if there were a good use for it. And if they don't pass the Turing Test right now, you can bet they will soon.
Any form of what we'd call AI currently "learns". It's just a matter of when and what it learns, and whether there's a feedback mechanism.
We don't seem to be anywhere near general artificial intelligence, that can apply itself to any domain. I'm sure the new planes are awesome at flight physics, but I'd wager they can't tell you what to substitute for eggs in the cake you want to bake.
I have been following computer science and AI for decades now and, through observation, have come to the conclusion that:
"AI is that which has not yet been developed - anything else is just a computer program".
At least, that was true until a few years ago when popular media grabbed the term from academia and applied it to anything that showed more intelligence than the average journalist - a very low bar indeed🙈