Main image of article AI risks: What’s hype, what’s real and what’s next

AI has become an everyday item in society. Approximately 50% of the US population now uses chatbots and 20-25% use generative AI for creating images and more.  There are concerns that AI poses a risk to humanity, but it comes in different ways. Here are five possible types of threats.

Summary

Let’s dispel the science-fiction notion that AI is going to destroy us via rampaging terminators after a nuclear war. AIs do make mistakes (hallucinations), so giving them launch control of nuclear missiles would be a very stupid thing to do.  

But AI controlled weapons do exist, for example robot sentry machine-guns are used to guard areas near hot spots like the Korean Demilitarized zone. You can even buy the non-lethal version to protect your home; they fire paint balls instead to mark intruders!

Another example of AI controlled weaponry would be drones, both Ukraine and Russia claim they have used them to kill soldiers on the ground but the overall attitude to autonomous drones killing people is negative. Lethal Autonomous Weapons Systems may be banned from targeting humans, however are still utilized as anti-drone weapons.

However, AI does simplify the task of creating nuclear, chemical or biological weapons for smaller nations. A senior scientist at the National Institute of Standards and Technology (NIST) demonstrated via a peer-reviewed mathematical proof that no AI’s guardrails are impervious to the right combination of words because natural language has infinite ambiguity.

This doesn’t appear to have happened, rather there is job displacement with a reduction in entry levels jobs, but a rise in jobs such as prompt engineering. This is in line with previous technology shocks to employment. Between 1900 and the 1930s, automobiles displaced jobs in horse trades (farriers, blacksmiths, carriage makers and so on) while creating between 4 and 5 million new manufacturing and repair jobs. In the 1990s affordable PCs, word processors and spreadsheets wiped out roughly 2.8 million data-entry jobs, but nearly 4 million PC jobs were created.

What new jobs has AI created? Here’s a short list:  Prompt Engineers, AI Safety & Alignment Researchers, Synthetic Data Specialists, Responsible AI & Governance Professionals, AI Operations (AIOps) Specialists, AI Agent Designers & Orchestration Engineers, Big Data Specialists and Autonomous Vehicle Specialists. It’s only been a few years since generative AI has gained wide use, so expect a lot more jobs to come.

However, AI job displacement is also having an effect on some roles, such as Junior Software Engineering and Legal Paralegals.

This has happened and will continue to happen for some time to come. AI has exposed many of the vulnerabilities in existing software and infrastructure. It’s not all that surprising that they exist as the tools that we used are still based on programming languages designed 40 or 50 years ago— looking at C/C++ in particular.  Newer, safer languages like Rust are starting to make inroads in both Linux and Windows. Just as AI is good at finding vulnerabilities, it’s also good at helping to fix them.

If you think you can distinguish between real and AI generated images, take a look at the Which Face Is Real website. My score was close to 50%, no better than random guessing. We’re not quite there with video, but a time when it is impossible to determine if a video is real or AI generated can’t be far off.

When it comes to text, it’s an arms race between AI text generation and AI detectors.  There were various ways of detecting AI text, for example if it contained long em dashes, but as the various detections were discovered, workarounds were produced. Now, largely because of the EU AI Act, AI-generated text is required to have watermarks. This has been done by AI carefully choosing some words in its output to embed a detectable secret key that evades removal when a user is just copy and pasting. Anthropic, OpenAI and Google have all done this.

AI seems to be cropping up in more places, even if you don’t realize it. Netflix, for instance is updating its recommendation algorithm with an internally trained LLM called GenRec.  Online search is now effectively a chatbot. All this is projected to add over $15 trillion to the global economy by 2030 and has made the leading AI firms (aka Big Tech: Apple, Microsoft, Meta and Amazon) not only richer but more powerful.  A comparison that writer Joel Kotkin describes as “The relationship between the oligarchs and the political elite more closely resembles that between the barons and King John at Runnymede – although there’s no contemporary Magna Carta in the offing.”

Many perceive that Big Tech has long leaned politically toward the Democratic Party. Could AI be employed to shape political attitudes in the general population if it isn’t already?  No matter which way you vote, this is not a good thing. Perhaps it’s time for politicians to act in order to keep AI out of politics.

These are still the early days of AI. Achieving artificial general intelligence (AGI) would open a real can of worms; for instance, would AGIs be recognized as a race with rights? Donald Rumsfeld’s famous quote about ‘unknown unknowns’ seems quite appropriate for future AI risks.