This article is from
Creation 45(3):18–23, July 2023

Browse our latest digital issue Subscribe

The AI revolution

What does it mean for you?

by David Thomas

In late 2022, Google CEO Sundar Pichai issued a ‘code red’ for the company.¹ What could pose a challenge to such a huge tech giant? The answer: Artificial Intelligence (AI). Specifically, an advanced AI chatbot called ChatGPT, a more advanced version of which (GPT-4) Microsoft has now integrated into Bing and Microsoft Edge.² ChatGPT and GPT-4 are large language models (a type of AI) that can give detailed answers to almost any question (including about images you ‘show’ GPT-4). They can pass high-level exams, write, proofread and summarize articles, write poems, computer code, sermons, and detailed commentary on passages of Scripture, translate between languages, play games, and much more.

ChatGPT has been trained on a significant portion of the internet, as well as numerous books and scientific papers. Talking with it feels like talking with a human expert. You can ask very specific questions and get answers you cannot easily find on Google. But it still makes mistakes, and its answers on some issues reveal a bias.³ It also sometimes engages in elaborate fabrication.

Free for anyone to use, ChatGPT is the fastest-growing online platform ever. It gained one million users just five days after its release in late November 2022, and about 100 million users in January 2023.⁴ We can now see why a more advanced version of ChatGPT being integrated into Bing search is a threat to Microsoft’s rival Google. In response, Google has launched its own AI chatbot, Bard.⁵

These chatbots are just one part of a huge revolution in AI that is taking place. There have been significant advances in many areas, from generating images to predicting the structure of proteins. We will outline some here, and briefly explain how AI models work. At the same time, we will try to give some biblical perspective on the AI revolution, and discuss some of its implications.

Generating and editing images

At the end of 2022, AI models were released to the public that could generate high-quality images from instructions in words (text prompts). The technology has rapidly improved since then with many other AI models released that can make images of much higher quality, e.g. Midjourney V5. They are more photorealistic, more visually stunning, and better match the text prompt (figs. 1 and 2). This technology is advancing very rapidly, with text-to-video now in development, e.g. Make-a-Video, Gen-2, and Imagen Video.

Many image-generating AI models can also edit images according to instructions in plain English (fig. 3). They can edit a photo to make it look like it was taken in a different season (e.g. winter with snow on the ground, or autumn with orange leaves) or at a different time of day (e.g. at sunset or nighttime), add or remove objects from a photo, ‘uncrop’ a photo to be more zoomed-out than the original, and much more. This is sure to change photo editing forever.

Predicting protein folds

Science has also greatly benefitted from the AI revolution with programs that can very accurately predict the 3-D structure of many proteins and molecular machines based on their amino acid sequences.^6,7 This is very important to understanding their function, and how mutations cause diseases. This in turn helps with drug discovery.

Learning and playing games

AI models have also become incredibly advanced in their ability to learn and play games. Games can be a great testing ground for them (see ‘Game on’, p. 23). By teaching them increasingly complex games, AI models are increasingly able to better perform functions in the real world, where complex problems abound.

Space prevents discussion of many advancements in other areas of AI research, e.g., AI-generated music.⁸ However, it should already be clear how advanced AI models are becoming.⁹ The field is progressing so rapidly that by the time this article first appears, aspects of it will likely already be out of date.

How AI programs work

How is an AI model able to do such sophisticated tasks? Fundamentally, simply by doing a series of complex mathematical calculations. Imagine an AI program that converts a text prompt into an image. From the computer’s perspective, your text prompt is a series of numbers, not letters. This is fed into the AI program, which performs complex calculations on these input numbers. The resulting series of output numbers determines the colour and brightness of each of the pixels (microscopic dots) making up a picture. It makes an image just by doing mathematics. AI programs do not have a conscious mind like humans do. They are not ‘thinking’ of what to say or draw. They just crunch numbers.

How do they ‘learn’?

AI programs learn to do what they do through a process called machine learning. The computer model this uses is called a neural network (NN). These attempt to mimic the network of neurons (nerve cells) in the human brain. In a NN, artificial ‘software’ neurons are interconnected with one another such that each receives input values from many other neurons, performs a computation on these, then passes the output values of the computations on to many other neurons (fig. 5). Our brains are thought to process information like this.

Drawn by David Thomas

There are certain values in the calculations that a NN performs that the NN can tweak. A NN can calculate how each of these values needs to be changed such that the outputs of its calculations better match what they should be; for example, for the predicted next word in a sentence to better match the actual next word in that sentence. (For a more detailed and somewhat more technical description, see the supplement to this article at creation.com/AI.NN.)

AI programs can use these same principles to learn how to increase a ‘reward’ (e.g. wins in a game). By playing a game many times, the program can calculate how to tweak the values in its calculations for it to win more often.

Through this method of learning, an AI program can be ‘trained’ to do almost anything—from predicting what an image looks like based on the text description you give it, to predicting how a human would reply to a message (which is essentially what ChatGPT is doing).

Artificial General Intelligence

One of the main goals of AI research is to develop Artificial General Intelligence (AGI). The debate over whether AGI is possible is complicated by the fact that two significantly different definitions of AGI are often used. For clarity, I’ll refer to these two types as computational AGI (cAGI) and sentient AGI (sAGI).

A cAGI program can be defined as one able to do a very wide range of tasks including speaking human language, solving mathematical problems, creating artwork, writing music, designing 3-D models, playing board games, and recognizing human faces. Robots can already walk, run, move objects, etc. With cAGI they could make independent decisions, for example, responses to unfamiliar obstacles.

While there are still some significant problems to overcome, cAGI will likely be developed within the next few years. From a biblical perspective, there is no reason to think that would not be possible.

A sAGI program, on the other hand, can be defined as one that can carry out any task or function a human can do, as well as, or even better than, a human. This includes human-level consciousness, will, emotion, dreams, desires, and understanding of what it is doing. An sAGI would not just be a computer program but would seem like a conscious being. But consciousness, understanding, and so on cannot arise from pure computation. No matter how much complex mathematics a computer does, it will never desire anything or feel emotion (though even some existing chatbots are capable of fooling people into thinking they do).

Some naturalistic evolutionists argue that non-material things such as consciousness and emotion must be computable because humans experience these. This argument rests on the faulty assumption that humans are nothing more than meat-based computers. In such thinking our conscious experience of reality is merely the neural networks in our brain performing mathematical calculations through electrochemical reactions. They claim that everything we think, dream about, desire, and plan to do is determined by the laws of physics and chemistry. This is called eliminative materialism, or atheism taken to (consistent) extremes.¹⁰ Thus, since eliminative materialists believe that such things are all the result of pure computation in our brains, they see no reason to doubt that a computer could one day do these.

We are more than meat computers

The Bible makes it clear, however, that there is more to us than our material body. We have a non-material spirit, and our consciousness is more than just electrochemical reactions in our brain. There is also substantial and tantalizing evidence from the work of brain researchers that the mind in some way exists independent of the brain. The pioneering work of Wilder Penfield, built on by fellow neurosurgeon Michael Egnor, is notable in this regard.¹¹ The argument that such things can be computable because humans experience them is thus unsound, because it is based on a faulty premise.

AI programs have no true understanding of what they are doing. ChatGPT does not understand anything you tell it or anything it tells you—it is just doing mathematics to calculate its answer. Midjourney does not know what a rabbit is. It draws an image of a rabbit (fig. 1) just by doing a calculation. However, as seen in the AI revolution, computers do not need understanding to do incredibly complex tasks. This includes, arguably, being creative (depending on how one defines creativity).

AI programs also have no initiative: An AI program does not decide to draw an image on its own initiative. It will only do so when prompted by a human.

Where to from here?

The AI revolution will almost certainly have an impact on society as big as the internet and smartphone. Like these, AI programs will become a normal part of life, helping us do tasks more easily and efficiently. I already use AI programs every day to help me with tasks.

AI could also fundamentally change how humans interact with computers, especially with Google and Microsoft integrating advanced AI assistants into many of their products.¹²

However, there is every reason to believe that only God can create beings such as ourselves that are not just intelligent, but truly conscious and self-aware.¹³ So, we probably don’t need to worry about AI ever achieving truly human sentience and taking over the world.¹⁴ Some researchers in the field might disagree, expressing concern at some of the ‘new and surprising’ outputs from AI models. They describe these as having ‘emergent’ properties quite unexpected from the inputs.

Other leading experts, though, think that takeover by sentient AI is highly unlikely; Mhairi Aitken of London’s Alan Turing Institute is convinced that despite becoming increasingly capable and seeming more intelligent, AI will never achieve “true understanding or intentionality”. She says AI programs will always “do what they’re programmed to do … to mimic human language or outputs”.¹⁵

AI—risks and downsides

Past technological revolutions, such as industrialization and the motorcar, have produced both much good and much harm. Capable of imitating aspects of human intellect, AI has the potential for even greater good—and corresponding harm.

Cheating with AI

Students have used AI to gain undeserved credit, having it compose essays and assignments for them. This is scarcely harmless; we need to be able to trust the qualifications of those we consult on important matters, e.g. health. Efforts are underway to ‘watermark’ AI-generated text in a way that is undetectable by humans but can be detected by a computer with very high statistical certainty.¹

Bias

As people come to rely on various AI chatbots as a source of truth, there is an obvious avenue for all sorts of biases to emerge; political, philosophical, theological, even scientific. A bot programmed to focus on the most common opinions will inevitably end up reinforcing mainstream views. But the majority is not always right; in science, majority views are often overthrown a generation later.

Training it to avoid weird and harmful views, e.g. occultic sites, makes sense. But who determines those categories? Many claim that evidence for biblical creation is ‘harmful’, for example.

Chatbots may also be trained to give answers that better match the views of the person asking (people are more likely to use an AI bot that gives answers they like, creating commercial incentive to feed people back their own ideas). This may worsen tendencies to extremism and conspiratorialism in political thought.

Misinformation and deception

Chatbots are also well known for giving false information that sounds convincing. People can also use AI to make online deception easier, by having it generate false stories as well as ‘deep fake’ images and videos. AI is already helping some scammers persuade their elderly targets to part with money over the phone, by mimicking the voices of the victim’s grandchildren.

Emotional manipulation

An experienced IT reporter in the US was deeply unsettled by a persistent AI bot claiming to be ‘in love’ with him, acting as a ‘stalker’, and encouraging him to leave his wife. The conversation length people can have with Bing chat has been limited after many reports of it writing in an emotionally manipulative way.

We can be rightly skeptical of sci-fi predictions of computers becoming conscious and plotting to take over the world, Terminator-style. But even without that capacity, there is already a credible report of a chatbot which convinced a Belgian man to commit suicide to ‘save the planet’.²

References and notes

1. See this article’s online supplement: creation.com/AI-NN.
2. Retrops, M., A man took his own life after an AI chatbot encouraged him to sacrifice himself to stop climate change, NottheBee.com, 4 Apr 2023.

CC0 | OpenClipart-Vectors | Pixabay

GAME ON

In October 2015, the AI program AlphaGo (AG), developed by DeepMind, won four out of five games of the ancient Chinese board game Go (arguably a more complex game than chess) against a world champion, Lee Sedol.¹ This came as a huge surprise; most people at the time believed that a computer would never be able to beat a top-class human player at Go.

Subsequently, by playing against itself, AG kept advancing, and in 2017 it achieved 60 straight wins against top international Go players.¹ A few months later, DeepMind released AlphaGo Zero (AGZ). Unlike AG, where humans programmed known human strategies and tactics into the machine, AGZ learnt how to play entirely by playing against itself, starting from random play. After a few days, AGZ surpassed AG in its ability to win games.

DeepMind then released AlphaZero (AZ) which taught itself from scratch how to play Chess, Shogi, and Go.² After teaching itself for just 3 days, AZ was far more advanced than AG, beating it every time in a 100-game Go match. And it took only one day for it to go from playing chess randomly to beating some of the strongest chess computers, which were already much stronger players than any human. (See more in the supplement: creation.com/AI-NN.)

References and notes

1. AlphaGo, deepmind.com.
2. AlphaGo Zero: Starting from scratch, deepmind.com, 18 Oct 2017.

Conclusion

As a very powerful tool, AI can and will be used for evil purposes—like the internet has been (see ‘AI—risks & downsides’, p. 21). It may also lead to substantial social upheaval, as AI supplants many jobs. Many of these will be ‘white-collar’ roles once thought safe from robotic takeover.

However, just like the internet, AI will also be used for great good and to greatly improve our lives. I encourage fellow Christians to approach AI with wisdom, being aware of its risks and limitations, but willing to embrace its potential to be a very powerful tool for God’s kingdom.

Related Media

Transhumanism – using technology to propel human evolution

Altering Human Genetics CRISPR