How smart is AI really?

Today it is impossible to overlook the seismic shifts being powered by Artificial Intelligence (AI). The world is witnessing an AI renaissance, heralding a time of unprecedented advancements and unparalleled opportunities.
We would like to give you some insight in where AI development actually stands right now and how smart current AI solutions really are. The endgame is Artificial Super Intelligence but how far off are we?

The AI breakthrough

Recent developments in AI technologies have not only breached the barriers of imagination but have also set the course for a new era of innovation. From the advances in natural language processing that underpin this very text, to the breakthroughs in machine learning propelling us towards the tantalising promise of General Artificial Intelligence, we are in the throes of a transformation that is recalibrating our understanding of the world around us.

It's no longer about simply following predefined algorithms or performing repetitive tasks. Today's AI systems are able to learn, adapt, and even understand human emotions. They're decoding complex data patterns, making autonomous decisions, and evolving with every interaction. These developments are a testimony to the brilliant minds shaping this revolutionary technology, opening the floodgates to a multitude of opportunities.

These recent developments have led to very mixed reactions. On the one hand, many people are concerned about where this is going and how we can control AI. For example, a group of well-known tech leaders wrote an open letter last March to call for a 6-month pause on AI development. On the other hand, everyone has heard about companies letting people go and replace them by AI because of the immense potential.

AutoGPT is a revolutionary leap forward

One of the most impressive recent developments is definitely AutoGPT, built on OpenAI's GPT3.5 and GPt-4 technologies. Unlike its predecessors that required explicit user prompts, autoGPT is a remarkable "AI agent" that independently formulates and pursues high-level objectives, using various tools and the internet to achieve them.

Demonstrating advanced autonomy, AutoGPT generates and revises its own prompts in a recursive manner, working towards its larger goal without the need for continuous human input. Its capabilities also extend to managing memory, summarising to handle input length restrictions, executing unattended internet-based actions, and notably, writing, debugging, and testing code.

This advancement in multi-agent systems underlines the transformative potential of AI, positioning AutoGPT as a revolutionary leap that is reshaping the boundaries of what AI can achieve.

AI models miss a deeper understanding of the world

Current AI models have some well known limitations such as cutoff on training data (e.g.: GPT-4 is trained with data until September 2021), the inability to check facts and the inability to learn from user input. However the difficulty to reason about cause and effect is a particularly interesting one.

Current Generative AI models still generate responses based on patterns in training data. They try to predict the most appropriate answer instead of arriving at it via reasoning because they miss a deep understanding of the world. For instance, if asked why an apple falls down the tree, it might correctly say it's due to gravity because that is a common explanation in its training data. But if a more complex causality question is asked in a domain with less clear-cut answers in the training data, it may generate a less accurate or confident answer.

A great and easy example of this is DALL-E. The above image is generated by DALL-E based on the prompt "A kid who's sad because he dropped all his eggs on a concrete floor".

The kid indeed does look sad but the AI model does not understand why the kid is sad. It does not understand that those eggs should be broken.

Conclusion

In conclusion, we're in the middle of an AI revolution and we clearly evolved from single-purpose to generative Artificial Intelligence. We're definitely at a point where we can all hugely benefit from AI. From transforming businesses to streamlining our daily routines, AI has the potential to dramatically enhance our lives. So, there's no time to delay - the moment to embrace Artificial Intelligence is now. Nevertheless, it's important to remember that we're not (yet) at the point of Artificial Super Intelligence nor a world dominated by AI. The path to AI achieving such overarching influence is fraught with complexities and still requires a significant leap in knowledge. It's an uncertain, exciting future - one that's difficult to predict. So, let's dive in, stay informed, and be ready to navigate the fascinating journey that is the evolution of artificial intelligence.