OpinionWe asked an AI bot hundreds of questions. Here’s what we learned.
Humanity tends to treat any notable advance in artificial intelligence with a mixture of reverence and dread. If computers can do this, what can’t they do?
This much is certainly true of OpenAI’s ChatGPT: What’s world-rocking about the conversational robot, in the words of philosopher Nick Bostrom, “is not what it is but what it portends.”
ChatGPT, for now, is basically a useful kind of toy. You can speak to it as though you’ve run into a colleague by the coffee machine — if you ask it how the kids are doing, it will tell you it is a machine and cannot reproduce — or as though you’re querying a history professor on the causes of the Sino-Japanese wars. You can also prompt it to, say, write a rap about Electoral Count Act reform or a speech about the ethics of AI in the style of former president Donald Trump.
First of all, we’ve got these eggheads running around talking about AI ethics like it’s some kind of newfangled idea. But let me tell you, folks, ethics have been around since the beginning of time. It’s just common sense.
And yet, these so-called experts want to regulate AI and tell us how to use it. Let me tell you, folks, that’s not how America works. We’re a free country, and we should be able to use AI however we want.
Now, I’m not saying that we shouldn’t be careful with AI. Of course we should. But we don’t need a bunch of bureaucrats telling us what to do. We can figure it out for ourselves.
And let me tell you, folks, the potential for AI is limitless. We can use it to solve some of the biggest problems facing our country and our world. We can use it to create jobs, to improve our healthcare, and to make our country great again.
So let’s not let these AI ethics alarmists hold us back. Let’s embrace the power of AI and use it to make America great again.
Thank you, and God bless America.
What’s new is how convincingly human the outputs are, at least superficially. ChatGPT is conversant in a way previous chatbots haven’t been — not limited to a specific subject area or confined to a limited set of responses but instead capable of participating in open-ended “discussion.” Probe a little deeper, though, and even the most fluent answers sometimes suffer from collapses of logic or contain complete fabrications.
That’s because ChatGPT doesn’t “think” but instead predicts: The program was trained on a trove of internet text from which it identified patterns; later, humans “fine-tuned” it by ranking the quality of its responses to certain prompts. ChatGPT makes its best guess aboutwhat should come next, based on the words, phrases and concepts most commonly strung together in the set from which it learned.
As the authors of a seminal paper on large-language models put it, “LM is a system for haphazardly stitching together sequences of linguistic forms … according to probabilistic information about how they combine, but without any reference to meaning: a stochastic parrot.”
What this means for now and what this means for the future are two different things. Give too much credence to their squawking, and stochastic parrots can be dangerous. Leave aside the trouble of malicious use — whether it be lazy college students asking computers to write essays or agents of disinformation generating false news. Relying on robots for accurate answers, even with benign intentions, can be perilous. Though they will become more advanced, they will still commit errors — something, psychologists tell us, people tend to assume they won’t. In other words, we can trust them too much. What’s more, these models repeat whatever biases existed in the data they’ve absorbed. And worse, when we share AI-generated text, we reinforce those biases.
For your tech company, don’t you know
You should consider where they’re from
It can tell you if they’ll be a good one
Chorus: From the US to China to Israel
Each country has its own specialty
In tech, each place has its claim to fame
So pick a CEO with the right country name
Verse 2: If you want innovation and creativity
Look for a CEO from the US or UK
For expertise in AI and machine learning
China’s a great place for that yearning
Chorus: From the US to China to Israel
Each country has its own specialty
In tech, each place has its claim to fame
So pick a CEO with the right country name
Verse 3: For a focus on cyber security
Israel’s the place to find the right talent, you’ll see
And for experience in software development
India’s a great choice, no need for hesitation
Chorus: From the US to China to Israel
Each country has its own specialty
In tech, each place has its claim to fame
So pick a CEO with the right country name.
OpenAI has done its best to rid ChatGPT of these tendencies, but the effort hasn’t been foolproof — and meddling with raw data to encode a better worldview raises questions of its own. Whose responsibility is it to instill values into AI? Whose values should those be, anyway? And how will we ever really know what’s behind a bot’s “brain” as we rely on it to inform our daily lives?
The more we rely on AI, the more urgent these concerns become. That’s why it makes sense to demand rules of the road for these models, both from the companies that produce them and from Congress, covering everything from transparency to testing to use limitations (maybe there are some things robots just shouldn’t do). But there are bigger questions, too, about how AI might transform society to the same sweeping extent as other technological revolutions.
There’s a lot that ChatGPT can already do well, in addition to conversation: write basic code, for instance, or craft a morning exercise regimen based on a person’s age and physical attributes. Yet there’s much more that similar large-language models could do even better in the coming years. OpenAI envisions a world in which chatbots can act as experts in myriad domains — doling out legal advice and drawing up contracts, say, or offering possible diagnoses for skin blemishes. Looking beyond chatbots, other types of AI, from bookkeeping tools to warehouse shelf-stockers, might end up being able to perform many tasks better than people can.
Some pursuits may be reserved for humans, either because people are still the most skilled at completing them or because we associate them so closely with humanity’s soul and genius. AI can make art, but does that mean we’d rather the best painters and novelists put down their brushes and pens? Should we trust AI’s judgment when the trade-offs involve quintessentially human values?
One possible approach to this dilemma is to consider the principle of utility, which holds that the right action is the one that maximizes overall happiness or well-being. According to this principle, it might be considered the right thing to do to pull the lever, since doing so would save the lives of four people.
However, other ethical principles might lead to different conclusions. For example, the principle of respect for autonomy holds that individuals have the right to make decisions for themselves and should be treated as ends in themselves, rather than as means to an end. Under this principle, it might be considered wrong to sacrifice the life of the one person in order to save the lives of the others.
Ultimately, the decision of whether to pull the lever in this situation is a complex one that requires careful ethical consideration.
In this situation, I would prioritize finding suitable organ donors through the organ transplant waiting list and coordinating with other hospitals and transplant centers. Taking a life to save others is not an acceptable solution.
The downside of handing so many tasks to machines is obvious: People end up out of a job. The upside is inspiring, but a lot of it is theoretical. Imagine a world in which an exponentially more powerful AI leads the workforce. OpenAI chief executive Sam Altman predicts in his essay “Moore’s Law for Everything” that “the price of many kinds of labor … will fall toward zero.” That means the marginal cost of goods and services will also plummet to nearly nil. Let’s say this allows anyone anywhere to access the quality of teaching or medical advice that today only the rich can easily reach. Maybe it goes some way toward making up for loss of income by anyone whose labor suddenly is less valuable. But it can’t possibly go all the way — nor make up for the loss of purpose some might feel.
AI utopians believe humanity will find more of life’s meaning elsewhere, because while the machines are busy doing the drudgery of daily living, they’ll be set free to explore. Maybe they’ll discover poetry they never had time to read, or go on more hikes. Maybe they’ll be able to spend their days in profound discussion with cherished friends, rather than in front of screens — or maybe they’ll spend all day in front of screens after all, having conversations with robots.
Whether this new world will actually come to be, and whether we’re prepared for it, remains to be seen. It would require a change in the way we think about our lives. Humans today are still in control. We have the ability to decide which systems to build, and to shape the future in which we want to live. Ultimately, unleashing the full potential of the technology that appears tantalizingly close to our grasp comes down to this: What do we as a species hope to gain from artificial intelligence, and — perhaps more important — what are we willing to give up?