Can AI be playing Socrates?

Below, I summarize my opinions about AI, as my submission to the United States Office of Science and Technology Policy in response to the “Request for Information: National Priorities for Artificial Intelligence.” My sincere thanks to Shelly Palmer for bringing that request to my attention.

Can AI detect human misinformation?

I played with ChatGPT to get a sense of what it can do now and what could be next. I reported my experiences in two separate texts. “Politicians have already outsmarted AI” is about my testing to determine whether the current version of AI can help us find a better immigration policy.

ChatGPT repeated the argument of many pundits and politicians that foreigners cause our immigration problems, and it is practically impossible to do anything more than what has not been working for about a century. I expected a much more intelligent answer.

At the same time, I was impressed with the amount of relevant information that ChatGPT could intelligently pull from its database. In my second test, I sought how AI can evaluate the truthfulness of contradicting information pieces, “Can AI help Americans overcome the political divide?” It led to an intriguing question: Can AI be an unbiased moderator in political debates? It could be a path to overcome our deep political divide.

Presently, all extreme-view holders believe that the truth is on their side and their opponents are wrong because of their ideological biases and, kindly speaking, lack of wisdom. My humble human intelligence concluded that it is worse; almost all Americans are wrong on at least a few important issues. AI will have a steep uphill climb playing Socrates if I am right.

Will AI increase or reduce misinformation?

It depends on the political decisions we make now.

An article by two Harvard professors, Archon Fung and Lawrence Lessig, “How AI could take over elections — and undermine democracy,” voices typical concerns about the dangers of AI. The authors bring an example of AI applications that can spread falsehoods affecting elections. It is true. But it is not new; it was also true in the 18^th century with political pamphlets when printing became affordable. It was true with recent elections as well.

Former presidents Obama and Trump represent drastically different political concepts and governing styles. Both won elections thanks to using social media effectively. Both promised to end our ongoing immigration crisis, and both did not do it. Both pledged to reform our health care. Trump did nothing; the long-term benefits of Obamacare are at least controversial. That example debunks the article’s core argument that elections matter. They do not – at least in addressing the core issues, which define the well-being of the nation. Regardless of which politicians we elect, our essential problems remain unresolved. There are other political forces at play.

One may ask: Will AI make it worse? We can ask perversely, as well: Could it be worse? Or will AI play Socrates, bringing that fresh breeze that will turn things around?

Who is afraid of the truth?

AI will bring a rapid increase in productivity. Hence, its creators, like its “godfather,” Geoffrey Hinton, or the CEO of OpenAI, Sam Altman, alert us that many jobs might disappear and massive retraining might be needed. Also, bad actors might be among the earliest adopters. We should prepare for that.

But in any of their numerous interviews, I did not find a mention of the most significant benefit, which could be the biggest danger at the same time. My humble experiments showed that the current version of AI could not analyze the message in the sense of abstract thinking; it could not pinpoint obvious falsehoods or internal contradictions in logic. But Professor Hinton and Mr. Altman assure us that those abilities will come soon.

When this happens, AI can become an impersonal BS detector. It can improve efficiency in science and business, but it will unlikely be revolutionary because thorough scrutiny is a necessary survival skill in those areas. In politics, voters often prefer to hear what they like, even if it loosely correlates with the truth. Hence, twisting the truth is a survival skill of politicians. But when AI is able to judge the veracity or logical consistency of political statements, all hell can break loose.

Hence, when the leaders of the emerging artificial intelligence industry ask for government regulations, and the government enthusiastically jumps on the idea, anyone knowing a little about politics can deduct that whatever the government will do and regardless of how lofty the regulations sound, it will be about making sure that AI will not play Socrates, bursting the bubble of their baloney.

Can AI give us a period of unprecedented prosperity?

There is no excuse for the American elite’s inability to change our dysfunctional immigration policy. Canada has no immigration problem. Politicians will pull the wool over our eyes about troubles in Latin America but will stay mum that, in Canada, immigrants are about 23% of the population, while in the United States, it is only 13.6%. That almost 10% difference begs the question: Would we have any immigration problems if we had admitted about 33 million (10% of the current population) more legal immigrants within the past 20 years or so? American mainstream media do not raise that question. Why?

When post-Soviet nations joined the European Union in 2004, some Western Europeans worried that Polish workers would be coming en masse to take their jobs. Major countries like Germany and France imposed restrictions. Small Ireland did not. It had few immigrants before, so the anti-immigration faction was weak. The government saw an opportunity and opened the gates. Doomsayers expected an apocalypse. Within two years, Ireland’s population increased by 5%. After the dust settled, experts concluded that “it was the longest recorded period of continuous growth in the Irish economic history.”

Looking for ideas on how to deal with our immigration mess, Vice President Kamala Harris did not go to Ireland; she went to Honduras. In the US, 5% of the population is about 16 million. Even if all 10 million Hondurans would come to the US, we could still have a place for everyone from El Salvador, which has about 6 million people. No one in the White House, Congress, or the mainstream media sees the troubles in Latin America as potential blessings for the American economy.

What will be AI’s take on that issue? During an event in India, Sam Altman said that the ChatGPT algorithm could easily learn collective preferences when it comes to ethical choices. After more than a century of systematic human misinformation, the collective preference in the United States is to have as few immigrants as possible. In my trials, ChatGPT, as it is now, leaned toward that preference.

So much attention goes to the elusive misinformation that AI might bring, but no one notices that on immigration, humans created an iceberg of misinformation larger than the one that sunk the Titanic. If politicians, the custodians of those immigration falsehoods, have any say on AI, they will sink it like the Titanic.

AI can make all Americans 6% richer every year

Our health care policy is another inexcusable failure of the American elites. In 2019, 16.8% of the United States GDP went for health care. All other developed nations spent about 6% less and got better results. No doubt, the United States can get comparable results. It could be a lengthy conversation about why it is not happening. The problem is that no one in the political elite wants that conversation to be sincere. No one in the mainstream media even tries to go to the crux of the crisis so that informed voters could press their representatives for a workable solution.

Years back, when politicians pretended to discuss health care reform, I wrote my observations in an article, “Stone Age politics in the health care reform debate.” Nothing has changed since then. If politicians have any say in the AI implementation, they will make sure that AI will not deploy an artificial Socrates in examining our health care policy.

AI might disturb our understanding of climate change

When asked about climate change, the chief scientist at OpenAI, Ilya Sutskever, expressed hope that AI can help find technologies to capture CO₂ and lower its emissions. It is a view supported by many major media outlets. I challenged the Washington Post: “It is not about climate, and it is not change.” Interestingly, while avoiding taking sides, in my tests, ChatGPT was able to compile an argument supporting my point of view.

I looked closer at “Science and money in the climate change debate,” when alerted by proposals of lavish spending of my taxpayer’s money. Again, it does not matter whose arguments are closer to the truth. We all lose when public expenditures, measured in trillions of dollars, are decided by elbow pushers behind closed doors.

The money is in the truth

Every few generations, a nation must revisit its objectives and how to achieve them. The United States is at that moment now. It is time for historical reflection and a uniting vision. The limping economy we have over the last two decades is a measurable consequence of our political divide. The biggest untapped reserves in the American economy are in overcoming the entrenched obstacles to progress. Immigration, healthcare, and climate change are the most costly, but not the only, unresolved problems. Americans need intelligence only to resolve our problems. Human intelligence has failed us; AI is our last chance.

We do not need an authority telling us the “absolute” truth. We need a skeptical BS detector, objective and impersonal. The one that can accept the data without prejudice and can change its evaluation of the truthfulness of a given statement if the data supports it. Americans need Socrates on every computer and smartphone.

In the questionnaire, government officials ask how we can protect individuals and the nation from the potential dangers of AI. One cannot pass laws making Americans smarter. I elaborate on that observation in my polemic with Professor  Shoshana Zuboff, “Human and artificial intelligence in capitalism.” Now, when artificial intelligence can perform mental tasks that previously only humans could do, the government can unblock the debating of our problems. We could start with the three issues mentioned above. It might be tremendously troublesome in conventional media and politics. AI can make it doable.

We live in an information-driven era. People worldwide have the same talents and fragilities. But not all nations are equally prosperous. Those that can genuinely discuss their problems and make better decisions are wealthier than those that do not. The money is in the truth.

It means that the first 28 questions in the government’s request are not helpful. Answering the 29^th question, we need to bring Americans — including politicians and pundits — closer to the truth. Instead of talking about AI, the government should ask Americans to try it. The government should ask for opinions on fixing the three earlier listed problems. And it should ask OpenAI to create a Socrates-like BS detector to scrutinize the entries and engage in debates with the contributors.

After that trial, we can all talk about AI more intelligently. Of course, it assumes that politicians will not serve that artificial Socrates a cup of hemlock tea.