- Understanding the Potential Benefits of AI
- Understanding the Potential Risks of AI
- How to Spot AI in the Wild
This chapter is from the book
This chapter is from the book
This chapter is from the book
Understanding the Potential Risks of AI
Just as AI promises numerous potential benefits, it also comes with risk. Will AI really make your life better—or will it negatively impact your life and livelihood? Again, it all depends on what you do and how AI develops.
AI Can Spread Misinformation
AI is just a tool. It does what users ask it to do. If someone prompts an AI system to produce a picture of a giraffe flying a helicopter, AI will do it. The initiators of that prompt can then use that image in whatever way they want.
Thus, we come to a real-world issue concerning the use of AI content: mischievous or malicious individuals can use AI to create blatant falsehoods, either in words or visuals, and then spread those falsehoods over social media and other channels. AI-generated text, images, and videos can be particularly convincing, especially as AI models continue to improve. If someone wants to convince people that a giraffe can fly a helicopter, a photorealistic image of that scenario can be very persuasive.
Realistic but false text, images, audio, and videos are called deepfakes. In the past, people have used image editing programs such as Adobe Photoshop to manually manipulate images and create deepfakes. Today, AI image generators can do the job better and faster with just a few simple prompts.
Say, for example, you wanted to conduct a smear campaign against a neighbor you don’t particularly like. You can feed an AI image generator a picture of your neighbor and prompt it to create a photorealistic image of that neighbor burning trash on their lawn. You could then take this very real-looking picture to your neighborhood association and try to get your neighbor in trouble.
In this same fashion, celebrity deepfakes are popular. Take a popular actress, tell the AI image generator to create a picture of said actress sans clothing, and—voilà!—you have a ready-made pornographic image ready for distribution on the Internet.
This sort of AI-powered manipulation can also be used for political and propaganda purposes. We’ve already seen deepfake photos purporting to show people doing things they didn’t really do, deepfake videos purporting to show events that didn’t really happen, and deepfake phone calls impersonating politicians saying things they didn’t really say. AI can make these deepfakes extremely convincing—so convincing that voters could be swayed to change their vote from one candidate to another.
In addition, AI can be used to spread false information over social media. Most social media platforms use AI algorithms to decide what shows up in users’ feeds. Manipulate that algorithm just a bit, and you can fill peoples’ feeds with political falsehoods and biased viewpoints. That could be dangerous for any functioning democracy.
Even though the Federal Communications Commission (FCC) recently outlawed AI-generated political robocalls and major tech companies signed an accord to prevent AI from being used to disrupt elections, AI-powered misinformation remains a major short-term and long-term threat to elections in the United States and abroad. FBI Director Christopher Wray recently warned about foreign adversaries using AI technology to influence U.S. elections, saying that AI makes it “easier for both more and less-sophisticated foreign adversaries to engage in malign influence.”
If nothing else, the threat of AI-generated deepfakes could cause people to question even legitimate stories and images. If you can’t tell what’s real and what’s fake, what can you believe?
AI Can Be Biased
AI results are based on the data fed into large language models. The more data, the better the results.
Equally important to the quantity of data available is the quality of that data. In AI, as in most things, it’s a garbage in, garbage out type of situation. Bad data will result in unreliable AI models.
Because AI relies on the data it’s fed, biased data can be a significant problem. Remember, most AI models get data by scraping content from the Internet. Unfortunately, there’s a lot of flawed or biased content on the Internet, and those characteristics can be absorbed into an AI model.
Bias can infiltrate AI systems in a number of ways. AI can ingest training data that reflects historical or social prejudices. It can include training data that includes biased human decisions and use data that either over- or underrepresents specific groups, thus reinforcing existing biases. It can even treat opinions or obvious jokes in the training data the same as it does hard facts.
In addition, AI can exhibit the biases of the people who develop its algorithms. As AI researcher Olga Russakovsky notes, “AI researchers are primarily people who are male, who come from certain racial demographics, who grew up in high socioeconomic areas, primarily people without disabilities.” That creates a very specific worldview that is, to some extent, exhibited in AI output.
For all these reasons, AI content today often exhibits the same biases that exist in our society at large. Without conscious upfront programming, AI is likely to perpetuate those biases in the decisions it makes.
Consider, as an example, a company’s use of AI to vet job candidates. If an AI model reflects a society’s bias against specific ethnic or racial groups, a company will continue to hire fewer people from those groups.
Similarly, AI-powered speech recognition software can fail to understand certain accents and dialects because the generally white male researchers don’t speak that way. This can cause problems for an AI chatbot trying to understand or respond to questions from customers of certain backgrounds.
As another example, consider AI-generated images. Given the gender and race bias present in today’s AI systems, if you ask an AI tool to create an image of a businessperson, what are the odds that it will show a white male and not, perhaps, a female of color? (Pretty good; see Figure 2.1, the result of a single such prompt with DeepAI’s AI Image Generator.)
)
Figure 2.1 The stereotypical image generated by the prompt, photo of a businessperson. (Image generated by DeepAI’s AI Image Generator.)
AI models can also reinforce society’s historical age bias, especially in employment. This is particularly concerning given the use of AI-powered recruitment systems; if the algorithms used by these systems are biased toward younger candidates, older job candidates may be unfairly excluded from consideration for some jobs.
Guarding against all forms of bias is essential in creating trustworthy AI content. Otherwise, AI will increasingly exhibit those biases, both good and bad, that exist in our society today.
AI Can Invade Your Privacy
Here’s a major concern of AI critics today: AI is a major threat to our personal privacy, and they will only grow.
Think back to Chapter 1, “Artificial Intelligence: What It Is and How It Works.” Do you remember where and how AI models get all the data they use to train and learn?
The answer is that AI gets its data from you and me and everyone around us. Most of the data that’s been fed into AI large language models has been scraped off the public Internet. That means not only website content but also social media posts, online messages, and other communications between unsuspecting individuals.
That’s right, your friendly neighborhood AI model is based at least in part on your own thoughts and words, as well as those of your friends and neighbors. And the models have obtained your information without asking you permission. If it’s out there on the Internet, the thinking goes, it’s free for the taking. It might be different if the content is behind a paywall or on a private site that requires registration or permission. But anything that’s out there publicly, the AI companies say, is there for the taking.
That includes content that isn’t actually on the Internet but has been supplied online—conversations you’ve had with chatbots, chats you’ve had with friends and family, and questions you’ve asked on support sites.
To be fair, AI large language models don’t target your individual data per se; instead, they incorporate it and data from millions of other people into their models. It’s not using your data against you to do harm, as a scammer would, but it’s still using your data without your permission.
Despite numerous data privacy laws on the books, few if any such regulations explicitly protect your data privacy from AI. Some regulations have been proposed, but neither the United States nor the EU have enacted laws that cover AI’s use of personal data. That leaves your data out there for the taking by any AI large language model that wants to use it.
That’s not a good thing.
AI Will Replace Some Jobs
With every new technological age comes some degree of change and displacement. The industrial age eliminated many formerly manual factory jobs. The automotive age displaced workers in the horse and buggy industry. The rise of the Internet resulted in job losses in traditional media and communications companies. This sort of change is inevitable.
Don’t be surprised, then, if the biggest near-term impact of artificial intelligence technology is a significant loss of jobs. Companies large and small are looking to AI to help them improve their productivity, which means replacing expensive human workers with cheaper, less-demanding, more efficient AI systems. For these companies, “improved productivity” means fewer employees, which means layoffs—sometimes for even the most seasoned workers.
Now, many employers will couch this scenario as letting AI take over repetitive jobs so they can “repurpose” employees to higher-value tasks. While that is a possible scenario, it’s equally possible and perhaps more probable that many employees displaced by AI either won’t have the necessary skills for those higher-value jobs or that those jobs won’t exist at all. While the impact of AI will differ from company to company (and industry to industry), it’s likely to have a net negative impact on the human workforce.
What industries will be most impacted by the AI revolution? AI is likely to have an effect across the board, but in particular, anticipate job losses in the following sectors:
Agriculture, with AI-powered robots automating many manual tasks, such as planting and harvesting, especially on larger farms
Finance and banking, with AI automating both customer-facing and back-office jobs
Healthcare, with AI assisting or replacing many scheduling and back-office functions
Legal services, with AI taking over contract generation and management
Manufacturing, with AI-powered robots replacing factory workers
News media, with newspapers and websites using AI to generate articles and posts
Transportation, with self-driving vehicles eliminating human drivers to transform the trucking and rideshare industries
How big will this AI-powered job disruption be? Goldman Sachs estimates that generative AI could eventually replace up to 300 million jobs worldwide, with many occupations experiencing a 25 to 50 percent job loss. This would be a huge disruption to the job market—and to the way of life for hundreds of millions of workers.
Unlike the industrial revolution, which primarily impacted manual or blue-collar workers, the AI revolution is likely to also affect higher-paid white collar workers. That will be a major difference from previous technology-based changes and a big concern for skilled workers everywhere.
It’s not all doom and gloom, however. On the plus side, Goldman Sachs predicts that artists, computer system analysts, HR managers, legal professionals, mental health professionals, surgeons, teachers, writers, and those in leadership roles are less likely to be replaced by AI because of the need for human judgment and creativity in those roles. In addition, there is already a huge demand for jobs programming and training AI systems.
Will your job be one of those lost to AI? Perhaps, and even if not, many of your coworkers will be impacted. Prepare to be disrupted.
AI Will Make Mistakes
AI constantly makes mistakes—or in AI parlance, hallucinates. We all must verify its output. If we try to rely too much on AI, especially for mission-critical tasks, we will be disappointed when things go wrong, which they will. Witness AI-powered self-driving cars that get into accidents because of faulty or less-intelligent AI systems, or AI image generators that give people six fingers and a missing ear.
Today’s AI systems will make mistakes. Putting all our trust in said systems, at least at this point in time, is ill advised. If you rely completely on AI and AI isn’t perfect, the decisions you make based on that AI may be flawed. Likewise, if you’re using AI to manage operations or systems, you may experience system interruptions if AI gets some of the data wrong.
AI will get better and more reliable, but it’s not there yet, wishful thinking aside.
AI Uses Significant Resources
AI is a resource- and power-hungry technology. Today’s increasingly larger AI models require vast amounts of power, both in terms of electricity and computing power. AI models need many fast and powerful CPUs and GPUs, vast amounts of data storage capacity, fast and reliable Internet connections, and lots and lots of electricity to run it all.
Unfortunately, none of these items are cheap or limitless. AI is an expensive technology, which is why so many large models are the provenance of today’s large tech titans, such as Amazon and Microsoft, that have the financial and other means to pull it off.
Looking just at AI’s electricity needs, one expert calculated that by 2027, the AI sector will consume between 85 and 135 terawatt hours per year. To put that in perspective, that’s about half a percent of all global electricity consumption. That’s massive—and increasing daily.
At some point there may not be enough available resources to power all the AI systems currently being developed. What do AI companies do if there’s a chip shortage or a lack of storage or not enough electricity to go around? Or, equally likely, if the costs of these resources rise to unaffordable levels? The ability of the AI industry to grow may be constrained by resource availability and pricing.
In addition, all the resources that power AI have a major impact on the environment. The energy usage alone contributes significantly to fossil fuel usage and the resulting climate change. AI is not in the least bit environmentally friendly.
