If AI approximates “thinking”, then who are we? How does the evolution of AI affect human cognition, perception, and interpersonal interaction? The nascent age of AI obliges us to face whether there is a form of logic that humans have not mastered or cannot attain as it reaches realms of reality we have never known and may never know directly.

Powered by new algorithms and the cloud, humanity is developing digital architectures for exploring and organising reality. AI is accessing reality differently from the way humans access it and in ways that remain largely inscrutable to us. We are at the bend in the river, and during this AI transition there is a proliferation of non-human forms of logic with influence and acuity, in narrow settings for now, but which may soon exceed our own capabilities.

At this time, some of the outputs from the Third Generation Generative Pre-Trained Transformer (GPT) are prosaic repetitions and combinations of human ideas, others are hallucinations, and others are uncannily human. It is now clear that each program is mastering its subject differently from humans.

And in many cases, they have produced results that are beyond the capacity of the human mind to produce these results in the same time frame. In some instances, results were obtained by methods that humans can retrospectively probe and seek to understand.

In 2017, AlphaZero, an AI program developed by Google’s DeepMind, defeated Stockfish – the world’s most powerful chess platform. Before AlphaZero, programs relied on strategies and tactics that had been used by humans. The advantage they had was speed and processing power. No originality or imagination. Unable to understand sacrifice.

AlphaZero on the other hand was simply supplied with the rules of chess and a strategy to maximize its proportion of wins to losses. AlphaZero sacrificed its queen and executed many moves that humans had never considered. It had no strategy in a human sense but its style has led to further study of the game. What it had – was a logic of its own. It executed surprising tactics. It was unorthodox. The best players in the world must now – watch and learn.

In 2016, DeepMind Applied, a subdivision of DeepMind, developed an AI on many of the same principles of Alpha Zero to optimize the cooling of Google’s data centres. The best human engineers had already tackled the problem and produced great results.

DeepMind’s AI reduced energy expenditures by an additional 40 per cent – an improvement over human performance. As AI is applied in diverse fields and domains, the world will change. And the suggested solutions carry the watermark of nonhuman forms of knowing, learning, logical appraisal, and evaluation.

In 2020, MIT scientists discovered an antibiotic that was able to kill strains of bacteria that were resistant to every known treatment. Using AI, the MIT team developed a “training set” of two thousand known molecules. The “training set,” encoded data on atomic weights, types of bonds, characteristics of molecules, and the ability to inhibit bacterial growth.

From the “training set” the AI learned the properties of the molecules predicted to be antibacterial. But oddly, it also pinpointed attributes that had not specifically been encoded. Attributes that eluded scientists for decades. Attributes that once escaped conceptualization and human categorization.

Once training of the AI was completed, the AI was instructed to scan a library of 61,000 molecules, approved medicines, and natural molecules for molecules that would (1) be effective antibiotics, (2) not resemble any existing antibiotic, and (3) be nontoxic.

From the set of 61,000 one molecule fit the criteria. It was named “Halicin”. In the end, the AI did not routinely recapitulate conclusions derived from prior knowledge of molecular qualities.  Instead – it detected new molecular qualities. Fresh relationships between aspects of molecular structure and their antibiotic capacity that humans neither defined nor perceived.

Even stranger is that after the discovery, no one could precisely explain why the method yielded the result it did. The AI processed the data quickly, but it also detected aspects of reality that eluded humans, or perhaps those that remain beyond our ability to detect.

Shortly afterward, OpenAI demonstrated the AI called GPT-3 (“generative pre-trained transformer”). While AI does a particular task, GPT-3 generates responses to various inputs and cannot solve specific problems. Collectively, AlphaZero, the discovery of Halicin, and GPT-3 are the first steps in unveiling previously imperceptible aspects of reality that have great potential. As more software integrates AI and operates in ways that humans did not directly create, or fully grasp, AI will remain a fluid augmenter of our capabilities and experiences that will both shape and learn from our experiences.

The US Air Force has adapted the principles of AlphaZero to a new AI.  The AI algorithm, called ARTUu, flew along with a human pilot on a U-2 Dragon Lady, performing tasks that would usually be done by a pilot.

The AI operated the radar system without direct human oversight. The nascent age of AI is progressively posing epochal challenges to our concept of reality. The capacity of AI to evolve, disrupt, and learn will transform how we experience reality.