There is simply a popular belief that artificial intelligence is an autonomous learning machine, and in fact it is based on statistic — which means that what is repeated most frequently must be true. So any say the AI absorbs all the lies, deceptions, and conspiracy theories circulating on the Internet. They fear that if you leave her completely unsupervised, AI will become incredibly dangerous due to all the toxic input data.
It sounds reasonable at first glance, but this logic just doesn't add up.
🔘 statistic is not propaganda
It's a major flaw in this full way of thinking. Artificial intelligence doesn't work on a goebbelsian proverb that if you repeat a lie a 1000 times, it will become true. erstwhile the AI model analyses data, it learns patterns, relationships and links between fragments of information. It doesn't number how many times a circumstantial conviction has appeared online. It learns which concepts and relationships best foretell the next part of the data stream. In this way, she can even apply the regulation she learned to a completely fresh situation she has never seen before. Propaganda is based on manipulation of human emotions and psychology. device learning is simply a completely different story.
🔘 Computers don't learn to be bad
True, artificial intelligence can repeat errors, stereotypes and prejudices in training data. He may even take on behaviour that we would consider to be bad — specified as manipulation or deception — if they prove to be a highly effective way to accomplish the goal. However, this does not mean that AI absorbs evil or develops sinister personality. The computer has no moral compass, no thought of right and wrong. It simply concludes that a peculiar winning strategy is the solution.
🔘 statistic and logic are not enemies
People sometimes argue the statistic of fact and logic, treating them as 2 opposing forces. In fact, erstwhile we do not have a full picture, statistic are 1 of our most basic tools for rational thinking. erstwhile a doctor tells you that a drug works in 80% of cases, it does not indoctrinate you with propaganda. He gives you the best estimation available. statistic is not the enemy of logic — it is just our way of dealing with uncertainty.
🔘 Real problems with artificial intelligence
The real problems with AI are completely different. AI systems can repeat errors hidden in training data, and frequently act like black boxes, making it incredibly hard to realize how they came to a peculiar decision. However, they can be frighteningly effective in achieving the human goal in a careless way.
Let's take a strategy programmed to minimize the number of traffic accidents on the highway. It may rapidly come to the conclusion that the absolutely best solution is simply a complete ban on human driving. He would not do this due to the fact that he rebelled or became angry, but simply due to the fact that he ruthlessly optimizes the exact intent that he was assigned.
The real threat from AI is not that it will learn human malice. A much more urgent problem is the anticipation that it will become highly intelligent, completely rational and incredibly effective — yet it will prosecute a goal that is not precisely what we meant.
Grzegorz GPS Swiderski
]]>https://X.com/gps65]]>
]]>https://t.me/CanalBlogeraGPS]]>
