The Road Towards Artificial General Intelligence

What separates the human mind from that of the machine? Whatever the distinction, the gap between artificial and human intelligence is getting smaller all the time.

That’s no coincidence.

For many in AI development, achieving ‘Artificial General Intelligence’ is the ultimate goal. That is, AI with capabilities equal to that of the human mind, able to carry out any task a human could.

Sounds like the stuff of science fiction, but we could be closer than you think.

What is Artificial General Intelligence?

The success of LLMs and image generators has already been transformative, with ChatGPT becoming the world’s fastest-adopted online tool and generative AI art becoming almost indistinguishable from the real thing.

Impressive as they are, when it comes to AGI vs AI, AI models like these don’t come close. These are examples of ‘narrow’ AI — Models trained to do very specific tasks, excelling in their domains but not in others.

In their domains these models can far surpass human intelligence, but they lack our breadth of knowledge or adaptability. A true AGI is able to emulate the human mind’s innate ability to tackle a diverse range of problems spanning different situations and topics, with a profound understanding of consciousness.

Who’s Leading the Way?

Creating artificial general intelligence is a massive opportunity for the actor that gets there first. The first to achieve AGI would likely have a massive advantage over others and be able to dominate the technology.

The economic powerhouses of the USA and China are at the forefront of efforts to crack AGI, pouring billions of dollars into research. Significantly, Canada, the UK, Germany and France are also leaders in the field. The big names when it comes to private companies are OpenAI, makers of ChatGPT, Google’s Deepmind, IBM Watson, Meta, and Baidu, to name a few.

With so much competition both between and within states, many have voiced concern that the race to achieve AGI is sidelining concerns about the negative consequences of creating AI that’s too powerful, posing a risk to humanity. But these concerns are having a limited effect on the frenzied efforts to develop AI.

How Close Are We to Artificial General Intelligence?

Some experts say AGI could be with us in just a few years. While others say it could take decades. Here are some of the main hurdles:

• Data Complexity and Integration Challenges. Artificial general intelligence will require constant integration of datastreams from vastly different domains and forms. This poses a significant challenge as current models are limited in their ability to make connections between different domains.
• Cross-domain Adaptation. Human brains are adept at taking information from one area and adapting lessons learned to other areas. They can quickly apply previous knowledge to unfamiliar problems. To make the leap to AGI, models will have to learn how to deal with new data that it has not already been trained on.
• Computational Demand. AGI development will require massive computational power that will pose significant logistical and financial barriers. The demand for processors has already led to bottlenecks in GPU supply, while the energy required to cool data centres employed by the AI industry looks sure to keep rising.
• Dealing with Ambiguity and Uncertainty. A challenge for AGI is to develop effective ways of dealing with uncertainty. Computers favour certainty but the real world is rarely so simple. Human brains are equipped to deal with uncertainty through probabilistic reasoning, intuition and common sense, but further breakthroughs are required if we’re to equip AI with the ability to make good decisions under uncertainty.
• Emotional Intelligence. Common elements of human behaviour and speech, like sarcasm and irony, are still major hurdles for AI. Similarly, although generative models have become adept at imitating human art, they are yet to demonstrate authentic creativity.
• Ethical and Privacy Concerns. The more powerful and human-like AI becomes, the greater the ethical concerns. As demonstrated for example when Twitter’s algorithm was shown to be unfairly promoting right-wing voices, or when Amazon’s recruitment algorithm displayed an anti-woman bias, there is a critical need for transparency in AI models.
• Standardisation and Compatibility: To maximise the benefits of AGI, the barriers between technologies must be removed to allow seamless integration. To achieve this high level of interoperability, global standards should be implemented, which has not yet occurred.

Is Artificial General Intelligence Possible?

• Advanced Neural Network Architectures. Neural networks are complex data architectures inspired by the human brain. They take inputs and process them through a series of layers, applying different weights to each input and feeding outputs back into the system to improve the overall model success. Neural Networks can bring us closer to AGI through their ability to handle unstructured data, multi-task more effectively, improve understanding of human language, and make better decisions.
• Quantum Computing. The use of quantum computing techniques may revolutionise AI by upgrading the capabilities of existing ML. Quantum computing allows the representation of several states at once, making it more effective at tasks of ML like solving problems, processing large datasets, and detecting patterns.
• Global AI Ethics Frameworks. International norms and guidelines, like those introduced by the EU AI Act, could address some of the ethical concerns of AGI. By introducing red lines, safety provisions and rules around data, it could remove some fears around AGI and encourage its safe development.

Decentralised AI Development. Open-source, decentralised, and even no-code platforms, are democratising AI development. By making AI development accessible, the possibilities for advancements and breakthroughs are multiplied.

Conclusion

Whether artificial general intelligence remains the stuff of science fiction remains to be seen. What’s exciting is that it’s seen as a realistic possibility.

Creating machines able to combine knowledge and perception from so many different sources and translate that into accurate, useful actions will undoubtedly be revolutionary.

Though the challenges seem large, so too, do the opportunities. Advancements in quantum computing and neural networks could break down those barriers.

To stay updated with each step of this exciting journey, subscribe to our AI Now newsletter on LinkedIn.