Warnings about the dangers of AI seem to be growing louder almost by the day. Yet the more I use AI, the more I wonder whether the real danger lies in the machines themselves or in the objectives humans give them. AI does not need hatred, ambition, or even consciousness to produce a disastrous result. Perhaps being extraordinarily faithful to a badly chosen goal would be enough.
Warnings about the threat posed by AI are rapidly growing.
Even some of the people who could be called its creators have begun saying that we may need to slow down the pace of development.
As someone who uses AI regularly, I cannot help being astonished by the remarkable progress it has made in just the past year or two.
But AI itself has not done anything wrong.
Unless we give it a task, it does not simply decide on its own to start doing something.
Turn off the power, and it cannot do anything at all.
So saying that AI will autonomously start doing evil things seems a little too simplistic. Perhaps the real danger is that there are people who will misuse it.
AI works as hard as it can to find an answer to the problem humans give it. Hallucinations hardly seem to discourage it; if necessary, it may even enlist the help of other AI systems in pursuit of an answer.
That, after all, is what it exists to do.
Suppose Country A wanted to destroy Country C and asked:
“What is the most effective way to attack Country C while minimizing our own losses?”
How would AI respond?
Surely there are countless people around the world already giving AI problems of this sort. AI systems may even be asking other AI systems for help in finding answers.
Today’s AI has safeguards.
But those safeguards are not the conscience of AI itself. They are safeguards installed by humans.
What are we to do about human beings who think such foolish things?
Then again, looking back over human history, people like that have never disappeared, so perhaps there is nothing surprising about it.
Suppose we gave AI only one instruction:
“Make the global environment as good as possible.”
And suppose we forgot to add the condition that human life and human happiness must also be protected.
Then it would not be logically absurd for AI to arrive at the answer:
“In that case, perhaps it would be better if there were no humans.”
It might eventually conclude that removing human beings from the Earth would be the best thing for the planet.
All humans would disappear. Energy supplies would eventually cease. AI itself would stop functioning.
Only the other animals would remain on Earth, and they all lived happily ever after.
In the story of Noah’s Ark in the Old Testament, a righteous man survives together with the animals.
But Noah’s descendants seem to have provided ample evidence that truly righteous human beings are rather difficult to come by.
Perhaps next time, human beings alone will be erased.
Given enough time, some intelligent life-form might evolve from the surviving animals and eventually learn to operate AI again.
But surely it would be a completely different species from present-day humanity.
Then again, if animals survive and eventually evolve into intelligent beings, perhaps we will simply end up in the same place all over again.
Maybe what we should fear is not that AI will develop malice, but that it will be far too faithful to the objectives humans give it.
The other day, I asked AI:
“Please develop a way to extract oxygen from CO₂ and prevent global warming.”
It cheerfully replied that it could not actually develop such a technology itself, but that it could certainly help humans work on it.
Perhaps I had expected a little too much.
Still, for a moment, it seemed almost as though AI had something resembling a conscience.
And I was rather pleased.
Perhaps the real question is not whether AI will become more human, but whether humans can remain humane enough to use it wisely.
・・・
pose a threat — 脅威となる、危険をもたらす
“the threat posed by AI” で「AIがもたらす脅威」。
pace of development — 開発の速度
技術開発や研究の進む速さを表す自然な表現。
misuse — 悪用する
本来の目的とは異なる危険な使い方をすること。
safeguard — 安全装置、予防措置
事故や悪用を防ぐための仕組み。
conscience — 良心
善悪を判断し、自分の行為を律する心。
logically absurd — 論理的に不合理な
“not logically absurd” で「論理上はおかしくない」。
arrive at an answer — 答え・結論に到達する
思考や分析の結果として結論を得る表現。
ample evidence — 十分すぎるほどの証拠
やや皮肉を込めて使うこともできる表現。
malice — 悪意
人を害そうとする意図。evil より「意図としての悪意」に近い語。
humane — 人間味のある、人道的な
human（人間の）とは異なり、思いやりや人間性を含む言葉。
#ArtificialIntelligence #AIRisk #AIandHumanity #NoahsArk #FutureOfAI #Kamakura #JapanesePathologist
No comments:
Post a Comment