At the interactive science museum there was a section dedicated to various instances of artificial intelligence, with one being a very simple human against AI guessing game race. The human player could hit a left or right button, and the AI would try to predict which it would be. If it guessed correctly, the robot would move a click on the track, if incorrectly, the human would move a click.
The AI has won 70% of the ~150K races so far.
And I suspect it will increase that gap, as it uses historical data to train itself to better predict future actions of human players. However, it doesn't really need to, which is why this simple game is indicative of our own futures in an AI-dominated world.
There is still an expectation that AI has to be perfect to replace us. Or even, it has to be better than the best humans. But this is not the case. All it has to do to add value, is outperform the average human at a particular task. But the value added isn't one for one based on the task alone, because unlike a human, an AI can work continuously with no sleep or breaks, never gets sick, never goes on strike, never asks for a pay increase, and performs consistently and predictably, never suffering the slings and arrows of human existence. An AI doesn't call in sick on a Monday, nursing a hangover. And it is never affected by lack of sleep due to a crying baby, or struggles through a divorce.
Humans are fragile.
And not only that, despite our best intentions, we don't actually do a very good job learning from our mistakes, and worse at learning from the mistakes of others. An AI on the other hand will indeed learn from not only its mistakes and the mistakes of others, but also all of the success data points also. And rather than taking 12 to 18 months to make any impactful changes in its behaviour, it can include the new information into its code immediately, and start trialling instantly, in continuous improvement cycle.
I have no idea of the AI model being used in the game mentioned, but it would be interesting to know if it is a self-improving model. Essentially, it should really be a 50/50 guess, but if it is learning the patterns as it keeps getting played, I wonder how predictive it can get of human behaviours. My daughter was trying a lot of "random" patterns in the hope to beat it, but the AI would beat her, as it continually guessed more right than wrong. A 70% win-rate in a 50/50 choice game is incredibly high over so many thousands of games.
Humans are not only going to fall behind, but continually fall behind at an increasing rate, and AI and automation, including robotics, will advance and encroach into more and more human held positions. When AIs can predict a simple coinflip human better than 50%, it also means that they will be able to apply the same kinds of strategies to predict behaviours that rely on more complex data at a far better accuracy than a human equivalent. For instance, a psychologist AI would be able to combine a whole range of data points, combined with a host of cues like voice tone, eye movement, breathing speed and depth, heart rate and facial expressions, to give a far deeper appraisal than a human psychologist, who is only as good as their training, and awareness in the moment - and they might have been out drinking last night, or up with a crying baby.
To err is human.
But how much "error" are we willing to tolerate from humans, if there is a more accurate answer coming from AI? Do you want a human radiologist to read your scans, or an AI? Do you want a human financial advisor, or an AI? Do you want a human teacher for your children, or an AI? They sound like silly questions, but in fields that are seeing both shortages of staff due to being underpaid, and those that rely on processing a huge amount of data, or looking at very subtle indicators, the AIs are going to excel.
Think about the failing education systems where talented and engaged teachers are increasingly rare due to the salary and the declining quality of child behaviour, and the increase of home schooling. Some home school parents might have followed a trend, but not actually had the skills or process to do a good job of home schooling. However, with AI support, it is possible to have all the lessons packaged and tailored, possibly with teachers that will keep adjusting the style to meet the specific and changing needs of each individual. Some kids just aren't cut out for the normal school teaching styles, will AI help them learn above and beyond?
Learn for what purpose though?
What is the point of learning skills that are not going to lead to anywhere near a high enough level to compete with the skills provided by the AIs? What will the employment landscape look like for my daughter in two decades from now? What will her first jobs be like in a decade from now, when already the fast food jobs I did, are disappearing.
Where will she spend her skills?
Using skill for the fun of it only goes so far. It is great for a hobby, but when there are only hobbies available and nothing that is value adding to society, do we satisfy our natural human needs of contribution to the community? Already we are seeing young people struggling to find meaning in the world, with quite a few actually turning back to the religions for answers. It is like we have regressed as a society and instead of finding meaning in understanding reality, we are finding meaning by creating a fantasy again.
Maybe this will lead to a new age of religion where that mechanism in us is utilised to search even deeper than we can through science alone. But I suspect that it will instead be used to avoid the void - to fill up the space with something, so as not to face the reality of having nothing of value to do.
Our ultimate calling?
Or the final call for our species?
Taraz
[ Gen1: Hive ]