The Evolution of AI: Unleashing Prediction and Imagination
What is Artificial Intelligence?
In short, it is the science of teaching machines to learn. Rather than handcrafting a set of rules (“if this, then do that”), we want to get an AI system to learn a representation of what is useful or productive in some environment itself.
What is a Large Language Model?
It is a category of AI, specifically a vast deep learning network which has been training on many billions of words on the open internet. A prominent application of such models is sentence completion, wherein you provide an initial input, and the model predicts the most likely succeeding word or sequence. It is the dominant approach which has taken over in the last 3–4 years or so, with the GPT series by OpenAI and Google’s Bard being prime examples which all harness large language models or various flavours of the same infrastructure.
The Next Wave
AI technologies have the potential to take real-world actions. The previous wave of AI primarily centred around classification. Over the past 10–15 years, the focus has been on deep learning systems mastering the art of object and image recognition, audio transcription, and language comprehension. This process fundamentally revolves around classification, organisation, and structuring — creating meaningful categories.
Now, armed with the proficiency to classify effectively, these AI models can advance beyond and into the realm of prediction. Prediction is a prerequisite to taking actions, as before doing something in the world you imagine, you simulate, you plan. For instance, before going to the train station you lay out the route in your mind and make predictions on what the weather is going to be like, is the route going to be busy and are you going to get there in time?
This sequence of activities underpins human creativity, allowing us to envisage a whole range of possible outcomes. This capability to imagine these scenarios and subsequently select a course of action, ultimately enabling us to engage with and influence our environment — that is intelligence in action.
Now experts are trying to infuse this intelligence into algorithmic constructs and if this becomes accessible to many people it will bring in the MOST productive period in human history. It would enable everybody to take actions and intervene in the world in very new and novel ways. Especially in terms of creativity and invention, with millions of individuals capable of creating new innovations.
But of course, the flipside is, if you design those systems to have certain capabilities like autonomy or the ability to self-modify goals and act independently of human beings, that’s when they could potentially be very harmful or dangerous.
“Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”
— Centre for AI Safety
Containment
If you don’t start from a position of fear, then you probably aren’t paying attention. We find ourselves at the mercy of relentless exponential technological advancements. As technologies become more useful, they also become easier to make, more user-friendly, and cheaper to use, causing them to spread far and wide. This is extremely true in the context of software, which is unburdened by the traditional constraints of the physical world. Software can be easily moved around on the web; it is an idea, so it spreads even faster than a physical invention like a motorcar. If everything defaults to getting cheaper and easier, then we must contend with power in using AI to take actions in the world, being very widely available to many, many people.
Our intuitive understanding must keep pace with the rapid technological progress, and AI companies should willingly commit to subjecting their work to independent audits and sharing best practices. The European Union’s AI Act seems to be moving in the right direction. Their approach mandates AI developers to proactively assess the potential risks of their applications. If these risks surpass a certain threshold, developers must disclose the data upon which their AI systems were trained and how these models operate.
From a personal standpoint, it is imperative that we all embrace the idea of AI and its capabilities. The belief that AI is too complex to understand is nonsense, as AI is entirely accessible. Countless online videos offer a comprehensive understanding of AI systems within just a few hours of viewing. We should all strive to gain proficiency in AI because it is fundamental to its containment. To be effective stewards, we must grasp the fundamentals of what we are dealing with.
Artificial Intelligence, in all its forms, is a tool and one that will have countless applications tomorrow and in the future. If you want to find out more about what AI can do for you and your business, contact Version 1 today.
Bradley Clemson is a Java Developer at Version 1.
