Encyclopedia Britannica says that human intelligence is a “mental quality that consists of the abilities to learn from experience, adapt to new situations, understand and handle abstract concepts, and use knowledge to manipulate one’s environment.”
If one wants to believe what the media and politicians are saying, we now face the reality that computers are able to do the same. As usual, they sow fears that the machines will soon outsmart us. Villains will flood us with AI-generated falsehoods. They will cheat us out of our money and destroy the fragile American society.
While everybody is getting busy trying to protect us from the presumed dangers of AI, I wondered if AI could help us resolve some problems we cannot fix using our old-fashioned human intelligence.
Can AI help us find a better immigration policy?
Americans had perceived immigration as a problem since 1907, when Congress established a Dillingham Commission to study the case. Whatever the government did afterward did not work. Evidently, human intelligence failed us. Can AI help us find our mistakes and point us to a solution that would finally work?
I signed up for ChatGPT, the Open AI platform, bought a $20 monthly subscription for the premium service, and started chatting about immigration policy with the supposedly intelligent machine.
As an immigrant and a political writer who has been studying our immigration problems for about 20 years, I consider myself an expert on this subject. But in the back of my head, there always is a fear that I might have missed some important aspects of this issue despite reading almost everything essential. Hence, I was concerned that AI might temper my ego.
My opening question was broad:
“What is the best immigration policy?”
It was a little tricky because behind the “best” is a system of values. In my approach, which many do not share, the best policy increases the economic well-being of the host country the most.
I generated two responses about one month apart, having been curious if, in the meantime, ChatGPT might have learned something new. Later, ChatGPT confessed that the latest update it got was from September 2021. It does not have data beyond that point. Both answers to my opening question were like the mumbling of a slick politician. One could agree with everything written with round sentences and lofty words, but it was useless in advancing one’s understanding of the problem.
I followed with a challenging statement a few paragraphs long:
“Both answers are vague and useless. They say what the high school teacher would tell me long before we had computers, that the issue is complex and requires a lot of research.
“Let me pinpoint the weaknesses of the response I received today. In the introduction, it says, ‘The ideal immigration policy will depend on the specific goals and values of a country.’ There is only one goal of any policy that a government might introduce, bettering the prosperity of the nation. The problem is that politicians, experts, and advocacy groups might disagree about which policy will increase prosperity the most. Can ChatGPT help to sort out valuable ideas from that noise?
“To make things simpler, let us focus on finding the best immigration policy in the United States. Seeking expert advice or engaging stakeholders, as ChatGPT recommends, is precisely what concerned citizens, politicians, or journalists want to avoid. Each of them is aware that among the conflicting opinions circulating in the media, only some can withstand rigorous examination. Can ChatGPT help with conducting that scrutiny?”
Here is the full text of the ChatGPT answer:
I understand your concern about obtaining a clear and specific answer about what the best immigration policy for the United States might be. I agree that the ultimate goal is to better the prosperity of the nation, but different stakeholders may have different definitions of what "prosperity" entails and how best to achieve it. For example, some may argue that prosperity is best served by allowing more immigrants into the country to stimulate the economy, while others may argue that tighter immigration controls are needed to protect jobs for existing citizens. To help sort through these conflicting views, it can be helpful to examine empirical evidence from research studies on the impact of various immigration policies. I can't conduct new research or analyze unpublished data, but I can certainly help you interpret the findings from existing studies. Here are several key factors often considered in immigration policy debates, along with summaries of relevant research findings:
Economic Impact: Many studies find that immigrants, especially highly skilled ones, contribute positively to the U.S. economy. For instance, a National Academies of Sciences, Engineering, and Medicine report suggests that immigration has an overall positive impact on long-run economic growth in the U.S. They can contribute to innovation, start new businesses, and fill labor market needs.
Labor Market Effects: The impact of immigration on the wages and employment of native-born workers is a contentious issue. Some research suggests minimal impact, while others indicate certain low-skilled sectors may be adversely affected. For example, a paper by George Borjas suggests that immigration can have negative effects on the wages of native-born workers competing in the same job markets. However, Giovanni Peri finds that immigrants often complement the skills of native-born workers, raising their productivity and wages.
Public Services and Fiscal Impact:Immigrants use public services like healthcare and education, which has fiscal implications. A report from the National Academies suggests that while first-generation immigrants cost more in terms of public services, second and third generations provide a net positive fiscal impact.
Social and Cultural Impact: Some believe that immigration enriches the nation's cultural diversity and strengthens social cohesion, while others worry about potential societal conflicts due to cultural differences. These impacts are hard to measure and can vary greatly depending on local contexts.
Security Concerns: Critics often raise concerns about potential security risks associated with immigration. Empirical research, however, often suggests a low correlation between immigration and crime rates.
Please remember these are summaries and interpretations of complex studies and may not capture all nuances. Also, new studies may provide different perspectives and findings. Therefore, the best immigration policy might involve a combination of approaches, informed by continual study and responsive to the changing context and needs of the country.
In points 2-5, I recognize the DNA of propaganda by the most vicious anti-immigration advocacy groups, the same people who ran immigration policy during Trump’s presidency. For example, they emphasize a one-line mention in one of the George Borjas papers that after the chaotic arrival of Cuban refugees during the Mariel boatlift, the income of low-skilled workers slightly declined for a short period. They do not mention that everyone else benefited; from a historical perspective, it economically helped Miami. Professor David Card received the Nobel Prize for his study of the case. AI does not know that.
Similarly, illegal immigrants can be a burden on public services; legal immigrants much less, assuming that, as in Canada, they arrive when the job is ready. Again, AI unintelligently repeated the self-contradictory argument that the problems caused by our flawed immigration policy are a valid reason for not fixing it.
Regarding social and cultural impacts, AI repeats nativist propaganda by failing to detect that “potential social conflicts” only occur when “local contexts” are racist and xenophobic.
Lastly, on security risks, AI gives the same credentials to xenophobes claiming that there are some risks, and their opponents having data on hand proving otherwise. An eighth grader can spot the lack of intelligence in the AI’s take on this issue.
Continuing my chat, I wrote:
“The previous answer does not help to understand the immigration issue. Let me then ask a supporting question: How do we stop the steady inflow of illegal immigrants that has existed now for half a century? Canada is equally attractive. Why do they not have massive unlawful immigration?”
The ChatGPT’s response boils down to saying that we should continue what we have been doing:
Addressing illegal immigration is a complex issue and often involves considering a broad array of factors and solutions. One key to understanding the difference between the United States and Canada might lie in their respective geographic locations and economic disparities with their neighboring countries. The United States shares a border with Mexico, where significant economic disparities and violent conditions drive many people to seek a better life in the U.S. Also, due to the extended land border, it's physically more accessible for people to attempt to enter the U.S. illegally. Canada, on the other hand, is geographically more isolated and doesn't share a border with a country that has similar levels of economic disparity. Most of Canada's immigration comes from more orderly, legal channels because physical access is more challenging. However, this doesn't mean that Canada doesn't have illegal immigration; it's just a smaller-scale issue compared to the U.S. To address the question of how to stop the steady inflow of illegal immigrants, solutions often involve a combination of the following, although the effectiveness and moral implications of these measures can be subjects of intense debate:
Strengthen Border Security: This can include increasing the number of border patrol agents, employing more advanced surveillance technology, or constructing physical barriers. However, these measures can be costly and controversial.
Improve Legal Immigration Process: Making the legal immigration process more efficient and accessible can provide an alternative path for those who might otherwise attempt to immigrate illegally. This could include reforming visa systems, reducing application backlogs, or expanding paths to legal immigration.
Address Root Causes: Working with source countries to improve their economic conditions, enhance security, and uphold human rights can reduce the factors that drive people to emigrate. This approach could involve foreign aid, trade policies, or diplomatic engagement.
Workplace Enforcement: Strengthening the enforcement of laws against hiring illegal immigrants can reduce the "pull" factor of jobs. Measures might include mandatory employment eligibility verification and harsher penalties for employers who knowingly hire illegal immigrants.
Legalization Programs: Some argue for creating a path to citizenship for illegal immigrants who are already in the country, which they argue can bring these individuals out of the shadows, allow them to contribute more fully to society, and free up resources to focus on preventing future illegal immigration.
Cooperation with Transit and Source Countries: Cooperating with countries that immigrants pass through or originate from can help to manage migration before it reaches the destination country.
The choice among these options and their implementation details can be politically, socially, and ethically contentious. Each of them has potential benefits and downsides, and effectiveness can vary based on numerous factors. It's important to have nuanced discussions about these options, taking into consideration both empirical evidence and ethical considerations.
In Canada, immigrants are now about 23% of the population, while in the U.S., it is only about 13.6%, with about a quarter of them lacking legal status. That almost 10% difference begs the question that AI did not ask: If we had 30 million more legal immigrants (about 10% of the U.S. population), would we still have an illegal immigration problem? AI was not intelligent enough to spot that.
Realizing that it was going nowhere, I went into teasing mode:
“ChatGPT suggested that we can eliminate illegal immigration, among other measures, by eliminating root causes, meaning by that the internal problems in other countries. As this is a commonly expressed argument, I wrote an article explaining that instead of changing our faulty immigration policy, Americans blame others for illegal immigration. My reasoning contradicts the wisdom of ChatGPT. Where is my error?“
After telling me: “There is no error in your argument,” AI went with a long “however,” avoiding making any choices:
The AI council has evaluated your question from multiple perspectives and here is the consensus: There is no error in your argument that internal American problems contribute to illegal immigration. However, it's important to note that the issue is multifaceted and cannot be attributed to a single cause. Both internal American problems and root causes in other countries must be addressed to combat illegal immigration effectively. Here are some perspectives from the council:
Political Scientist: Recommends analyzing the push and pull factors that cause illegal immigration from various domains of politics and economics, not just within US borders.
Immigration Lawyer: Suggests that both internal American problems and root causes in other countries are interconnected and must be addressed together.
Sociologist: Emphasizes the importance of considering the impacts of historical and cultural factors on both internal American problems and issues in other countries that lead people to leave.
Economist: Believes that addressing only internal American problems is not enough to eliminate illegal immigration. Long-term solutions require addressing the root causes of economic instability and inequality in other countries.
Humanitarian: Urges the consideration of the safety, opportunity, and welfare of migrants, and not just focus on national interests.
In conclusion, while your argument about internal American problems contributing to illegal immigration is valid, it's also crucial to consider the root causes in other countries. A comprehensive and balanced approach that addresses both aspects can lead to more effective long-term solutions.
I became a little aggressive in my retort:
“The above response is what a slick politician would say when trying to avoid the question. Hence, let me add a supporting question. How many immigrants per year should the United States admit to reach both objectives, elimination of illegal immigration and boosting the economy?”
This way, I cornered AI to acknowledge that it cannot give any answer requiring “the abilities to learn from experience, adapt to new situations, understand and handle abstract concepts,” which defines intelligence:
I apologize for the inconvenience, but the available dataset does not provide sufficient information to determine the number of immigrants per year that the United States should admit to eliminate illegal immigration and boost the economy. The dataset focuses on migration statistics and does not include data on illegal immigration or economic indicators directly related to immigration policies. Determining the optimal number of immigrants for the United States to admit per year is a complex issue that involves many factors. These factors include but are not limited to the current state of the economy, labor market needs, humanitarian considerations, and the capacity to integrate and support new immigrants. It's also important to consider that immigration policies can have different impacts in the short term and the long term. This question might be better addressed by experts in immigration policy, economics, and social sciences who can consider all these factors and provide a well-informed estimate.
At that point, I started having fun, stating:
“The answer is wrong. The correct answer is: we should admit as many guest workers as they can find jobs in the United States. Those working for five years should automatically get the right to become permanent residents. Why does Chat GPT not know the correct answer?”
AI went into a defensive mode:
I apologize if my previous responses didn't meet your expectations. As an AI, I don't have opinions or beliefs, but I provide information based on the data I was trained on, which includes a diverse range of sources and perspectives. The solution you propose—admitting as many guest workers as can find jobs in the United States, and granting those who work for five years the right to become permanent residents—is indeed one approach to addressing the issue of immigration. It's based on the principle of matching labor demand with supply and providing a pathway to permanent residency. However, it's important to note that this is not the only perspective on this issue. Immigration policy is a complex and multifaceted topic that involves economic, social, political, and humanitarian considerations. Different stakeholders may have different views on the optimal approach, and these views can change over time and in different contexts. The role of an AI like me is to provide information and facilitate discussion, rather than to prescribe a single "correct" answer, especially on complex and contentious issues like immigration policy. I'm here to help users explore different perspectives and make informed decisions. I appreciate your understanding and patience.
I became more combative:
“Yes, ’Different stakeholders may have different views on the optimal approach, and these views can change over time and in different contexts.’ But it does not mean that all of those opinions are equally good. Humans are more likely to be biased than a data-based algorithm. So far, I see that human biases, particularly xenophobia, are incorporated into ChatGPT. The result is that, on immigration, Chat GPT advises doing with greater determination what humans failed to succeed within the past few decades.”
To my surprise, AI acknowledged its uselessness. Like a typical politician, it lied through its teeth that it made the best effort to be unbiased, but it also had to consider different stakeholders. In short, someone claiming that 2+2=5, if having some followers, is getting the same recognition as someone standing for easily verifiable truth:
I appreciate your feedback and understand your concerns. As an AI developed by OpenAI, I'm designed to be neutral and unbiased. However, I'm trained on a diverse range of data sources, which may include various human perspectives and biases. The goal is to provide a balanced view and not to endorse any particular viewpoint. Regarding the issue of immigration, it's indeed a complex and contentious topic with a wide range of perspectives. The approach you've suggested—admitting as many guest workers as can find jobs in the US, and granting those who work for five years the right to become permanent residents—is a valid perspective and could be an effective strategy in certain contexts. However, it's also important to consider other factors, such as the capacity of the host country to integrate and support new immigrants, the impact on wages and employment for existing workers, and the potential for exploitation of guest workers. These are some of the reasons why different stakeholders might have different views on the optimal pproach to immigration. Your feedback is valuable, and I'll strive to provide more balanced and nuanced responses in the future. Please feel free to ask if you have more questions or need further clarification on any topic.
Realizing that AI is still in its infancy, I asked:
“I challenged the correctness of ChatGPT responses about immigration and pointed out the strong bias favoring anti-immigration political forces in the United States. I wonder if my work on informing ChatGPT has brought any measurable improvements. I brought up many instances of publicly available and verifiable data during my chats. Can the algorithm learn from that data, or does it require human intervention to adopt new information?”
The answer surprised me:
I apologize for the confusion, but as an AI model developed by OpenAI, I don't have the ability to learn or improve from individual interactions. My responses are generated based on a mixture of licensed data, data created by human trainers, and publicly available data. I don't have the ability to access or retrieve personal data unless it has been shared with me in the course of our conversation. I am designed to respect user privacy and confidentiality. Your feedback is valuable and can be used by the developers to make updates and improvements to the model, but these changes aren't specific to individual users or conversations. The goal is to make the AI as useful and unbiased as possible for all users. I appreciate your understanding and your efforts to provide feedback and challenge biases. As of my current version (as of September 2021), I don't have the ability to learn or remember new information from individual interactions. I generate responses based on a static model that was trained on a diverse range of data sources, including books, websites, and other texts, up until my training cut-off in September 2021. I don't have the ability to access or learn from new data or information after this cut-off. This means that while I can generate responses based on the data I was trained on, I can't update my knowledge or learn from new information in real-time. Any updates or mprovements to my knowledge base would require a new round of training by the developers at OpenAI. This design is intentional to ensure user privacy and data security. I don't store personal data between conversations or use personal data to improve my responses. Each conversation with me doesn't have a memory of past conversations. I hope this clarifies how I work. If you have any more questions, feel free to ask!
AI is a mechanical writer
It seems that AI can write skillfully any text it is asked to write. But it is still far away from the ability to write intelligently, in the encyclopedic meaning of that term. In that sense, the term “artificial intelligence” is inadequate. The more proper would be a “mechanical writer,” a machine with encoded language rules, able to construct mindlessly linguistically correct text but unable to understand and handle abstract concepts.
People warning us about AI, seem to miss that part. An article by two concerned Harvard professors is an excellent example, “How AI could take over elections – and undermine democracy.” They are correct in explaining that modern AI applications can deceive voters. They do not see that as the newest version of the same misinformation that plagued print publications in the 18th century and the internet in recent years.
In every epoch, the proverbial sellers of the Brooklyn Bridge will find the buyers. It is not our biggest problem. Our inability to resolve our systemic problems – immigration is just one of them – is the biggest challenge. We have an ongoing political gridlock because of systemic misinformation caused by the lack of using our old-fashioned human intelligence where we can and should.
Human deception in American politics is still a few magnitudes better than the best AI can generate. So far, the AI algorithms are fed the same political mantra that dominates mainstream media. Politicians do not need to worry. The biggest weakness of AI is that it is still not intelligent enough to detect human-created misinformation.