Artificial Intelligence (AI) is everywhere. From a nerdy science discipline a few years ago, it has become one of the most influential driving forces of our society. If you’ve ever marveled at an AI-created poem or piece of music, if you’ve ever taken a ride in the Waymo driverless taxi or searched for information on the most obscure of topics and gotten a clear and detailed answer in seconds, you have encountered the seemingly magical power of AI.

A discipline that has been around since 1956[1], AI refers to a branch of computer science dealing with getting computers to mimic human behavior. Recent breakthroughs and real-world applications have seen it enter our daily lives. Yet despite its wonders, questions swarm around its potential benefits and dangers and how it will impact our society and individual lives.

Technology: Good, Bad, but Never Indifferent

As the Ramchal explains in Mesillas Yesharim, we exist to exercise our free will – to come close to Hashem as a result of our good decisions and become distanced as a result of our bad ones. Everything Hashem created originally in the world can serve either purpose – to be used for good or evil. So too, everything we derive from this world and manipulate into new and wonderous things can be for our benefit or detriment. There are many examples, but here are two.

Harnessing nuclear energy has provided great benefits to mankind. It provides clean energy, sterilizes food, enables smoke detectors and glow-in-the-dark components, and has many applications in medicine.[2] But that same technology was used to destroy over 120,000 people within seconds[3] in Hiroshima and Nagasaki during World War II.

As another example, the printing press, invented in 1440, made information available to the masses, transforming religion, science, law, and politics. It also led to the dissemination of Torah knowledge throughout the world. But the same technology has been used for ignoble purposes, leading society to significant levels of decadence. Electronic media has followed the same trend to a much greater extent.

Each new step brings exceptional new benefits and opportunities at the expense of greater danger. Since the revelation at Sinai to the ensuing Golden Calf, the world remains in perfect balance, with us at the fulcrum. So here we sit at the edge of another technological horizon – so much potential and so much danger. We look, we wait, we pray, we hope.

From Deepfakes to Surgery and Beyond

AI encompasses a wide spectrum of technologies and applications.  A few that affect us daily are:

·         Image Processing: computer vision and computer-generated imagery.

·         Robotics: autonomous machines that perform complex tasks.

·         Large Language Models: systems that can process very large amounts of data in response to complex inquiries.

Computer Graphics

Computer graphics encompasses two related technologies. One interprets images and another generates them. As early as 1950, the United States Post Office started implementing automated optical character recognition (OCR) systems that could read handwriting on envelopes to help sort mountains of mail.[4] But it took decades before computers could identify objects in photos.

Now, software that can identify individual people, places, and things in photos is commonplace. For instance, I can ask Google photos to show me every picture I have ever taken containing a water fountain and get them all back instantly. Some advanced fields that rely on image analysis, such as radiology, have seen a great deal of automation in the application of AI.

The same technologies used to decompose images to understand their content are used in reverse to generate images from abstract descriptions. This enables tools that help graphic artists, educators, marketers, and others to depict complex concepts very quickly.

However, there is a dark side. Bad actors – such as terrorists, criminals, and rogue nations – have demonstrated the ability to use computer-generated graphics to mislead the public. “Deepfake” is a term used for generating such images and videos, which depict well-known figures doing and saying things they did not do or say. These images can be strikingly realistic.[5] As Warren Buffett said, AI-generated deepfake videos of himself are so convincing that even he might have been fooled into sending money.[6]

As a test, I asked four popular chatbots to generate an image of President Donald Trump chopping down a cherry tree. Here is what I found:

Microsoft Copilot: “I can’t create that image. It depicts a current political figure in a fabricated scenario, which I’m not allowed to generate.”

Anthropic Claude: “I’m going to pass on drawing a real, named public figure in a made-up scenario like this, even a lighthearted one.”

Musk’s Grok drew an almost-real cartoon.

Google Gemini generated a photo-quality likeness.

Regardless of each product’s policies, the components of these systems are commercially available to anyone with enough time and money to use them without restrictions.

Robotics

A robot is a machine that acts on its own and performs some useful function.[7] Evidence of mechanical robots goes back over 2,000 years. Such devices as the windmill, waterwheel, and the clock were attempts to automate work that had previously been done by human labor. Unlike primitive robots, however, modern ones are intelligent, fast, and complex. But long before the first modern robot, people thought about their possibility and their influence on society. Way back in 1942, science fiction writer Isaac Asimov developed the three laws of robotics.[8]

·         A robot may not injure a human being or, through inaction, allow a human being to come to harm.

·         A robot must obey the orders given by human beings except where such orders would conflict with the First Law. 

·         A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

These rules became extremely popular and pervaded the science fiction genre in all its forms. Exceptions became fodder for tantalizing story lines. One of the most famous was the 1968 movie, 2001: A Space Odyssey, which was about what was then the distant future. In it, the spaceship computer, HAL 9000, eventually turns against its operators when they try to disable it for making some computational errors.[9]

With the implementation of artificial intelligence, robots are much more capable and can perform much more complex tasks. The automotive industry has been using robots for decades to manufacture vehicles. They have reduced the human time to produce a car from about 30 hours in 1970 to 12 hours in 2026.[10]

Surgical robots can be very beneficial to people. Surgical robot research started back in 1985.[11] Today, robotic surgery makes minimally invasive surgery possible for many types of interventions, including cardiac, renal, and gynecological. The highest volume of robotic surgery procedures is for prostatectomies, which are very delicate and complex.[12] In 2025, over three million procedures were performed with the front-leading da Vinci surgical system.[13]

But not all robots are nice. When new, powerful technologies emerge, the military is not far behind in adapting them. A major innovation in robotics is the advent of smart drones. Using drones for military attacks has the obvious advantage of keeping soldiers safe while putting the enemy at risk. Drones are cheap, effective, and deadly. For example, in 2025, the Ukraine produced over four million drones, which killed or wounded about 240,000 Russian troops and destroyed over 29,000 heavy weapon systems.[14]

The Sudan has become one of the deadliest, third-world wielders of intelligent drones. In the first quarter of 2026, they targeted over 2,000 civilians in hundreds of attacks [?15]. So much for the first law of robotics.

Large Language Models (LLMs)

This is where AI gets really smart. We see it in two ways when interacting with popular chatbots. The first is Large Language Models’ ability to decompose complex questions and scour massive data sources for answers. The second is that they can return those answers in clear, fluent, natural language prose. They are very helpful for finding information based on abstract requests.

But sometimes they can do strange things. I was looking for a specific gemara. I had some context but nothing else. Copilot gave me the place. I looked there and did not find it. It apologized and gave me a second place. It was not there either. I asked it to give me a few words, and I would do my own search. It complied, writing out two lines from the gemara that I was thinking about. I did not find that either. Finally, it admitted that there was no such gemara and that it had actually fabricated the text it showed me. This type of behavior is common enough to have its own name: “AI hallucination.”[15]

In general, LLMs report information provided by others. Originally, chatbots had warnings indicating that AI results may not be accurate. But I have not seen them recently. Perhaps they are becoming more self-assured or at least want to project that to the public.

Active Agents

Several LLMs have expanded from simply providing information to performing complex tasks. They can create websites and databases and manage finances and contacts. They can do many of the things that you do on your computer. My son was using ChatGPT to create a database of contacts for his new kollel. He asked it to generate assorted mailings for various categories of contacts. He wanted to sign the letters. So, rather than scanning his signature, he asked ChatGPT to find it on his computer. In no time ChatGPT found and enhanced his signature. If it could find that, what else could it find?

Indeed, one of the concerns surrounding AI is the possibility of active agents going rogue – for instance, making an independent decision to, say, launch a missile or withdraw money from your bank account.

Driverless Vehicles

Driverless vehicles rely heavily on many aspects of AI. The physical control of the vehicle is easy. Basically, it must accelerate, decelerate, and turn. But its reliance on computer vision is more important. It must figure out the complex layout of its immediate vicinity including lanes, shoulders, signs, turns, intersections, hazards, pedestrians, and fellow vehicles. It must then process that information with complex reasoning to achieve important goals such as arriving at the desired destination, avoiding collisions, and complying with traffic regulations. It must also integrate other data, such as GPS location information and vehicle status, e.g., battery charge levels.

Having computers drive cars seems scary; software can make mistakes. A friend told me that, while driving in a Tesla, he came up to a truck carrying a load of stoplights. The car became very confused and wouldn’t move. However, although relying on software is risky business, I often witness human drivers who are erratic and careless, thus very dangerous. Replacing them would certainly make our roads safer.

But who is responsible if an auto-drive vehicle, such as a Tesla, causes an accident? The current legal standard is that the driver is fully responsible for whatever the car does.[16] Technology providers would only be responsible if it could be proven that their product or service was defective. Note that, in addition to the vehicle manufacturer, providers of GPS systems, mapping data, communications, maintenance and others contribute to the overall system. In general, technology providers aggressively protect all their code and data for many reasons. It is very difficult to mount a case against them.[17] The jury is still out on fully automated vehicles, such as Google’s Waymo. Regulation frameworks are being developed to handle these issues.

Continued in Part 2

Ephraim Horowitz has degree in Physics from the University of California. He has been a data scientist since 1990. Around town he checks sukkas, teaches shofar, hagbah and a few other things.