Artificial Intelligence and the surprising strides toward AGI

14 April 2023
Paul McGillivray

In recent months, the technology landscape has witnessed a remarkable surge in progress. Interaction with Large Language Models (LLMs) via ChatGPT has become a part of many businesses' daily workflow.

Recently, groundbreaking experiments with intelligent agents have become the focal point of conversations everywhere, along with talk of the imminent emergence of AGI, or Artificial General Intelligence.

Artificial Intelligence (AI) and Machine Learning (ML) have been buzzwords for a while now and for a good reason. In technology, every new development remains in the realm of programmers, engineers and scientists until a level of abstraction is achieved that allows the general public to easily interact with it. Then it becomes mainstream.

Computers are, at their very deepest core, a series of electrical states of ‘on or off’. Initially simply switched, the microchip allows billions of these states to be set within a tiny space. The first computers were operated with punch cards to control these states. Then a keyboard with typed commands. These interfaces offered layers of abstraction that meant that humans didn’t need to manually change each state individually, and with the invention of the mouse and graphical user interface, there were enough accessible layers of abstraction for computers to be adopted into mainstream daily use.

ChatGPT was the level of abstraction that broke the daily use of AI into the business and creative worlds. Thanks to the work of OpenAI, a company Co-founded by Sam Altman and Elon Musk and invested heavily by Microsoft, the ability to use natural language processing to create text and images was released to the general public. The first available iteration was GPT-3. Many more tech-savvy people started using it, but it wasn’t until the introduction of the conversational interface of ChatGPT on top of the GPT-3 (and the subsequent more-powerful GPT-4) language model that its use exploded, alongside the explosion of Image generation tools powered by natural languages, such as DALL·E 2, Midjourney, and the open source Stable Diffusion.

We’re in a fast-moving landscape, now. RunwayML has revealed Gen-2 - a powerful Text-to-Video creative tool which can create impressive videos based purely on text prompts and also apply text and image commands to existing videos. The results are astounding.

Copilot for the Microsoft 365 suite promises to deliver everything you dreamed of when that awful paperclip character first popped up in Microsoft Word as the Office Assistant. In contrast to Clippy’s abject failure, Copilot will let you enter commands and requests in natural language; “Turn the transcript of last Wednesday’s Teams meeting into a Proposal”; “Take this year’s financial figures from Excel and pull out the key metrics”; “Create a graph in that same Excel document that shows projections for the next 12 months”; “Turn all of this into a Powerpoint presentation with graphics from my OneDrive”.

Those aren’t fantasy visions of the future - they’re abilities Copilot already has. The potential is both exciting and bewildering.

Another huge breakthrough in AI that will have ramifications for everyone is Autonomous AI Agents; software entities capable of acting independently, making decisions, and achieving goals without continuous human intervention.

Scientists revealed a lab study of Agents based on LLM’s. These Agents are autonomous pieces of code that appear to understand requests and conversations and - crucially - can define and then carry out tasks in order to achieve their goals. In a simulation of 25 Agents in a virtual environment similar to The Sims computer game, the agents conversed with each other, shared knowledge and ideas, and one even planned a Valentine's party which several other agents attended. (Here’s a good summary of the experiment and its implications).

AutoGPT - open-source Python application by Toran Bruce Richards -  allows AI to act autonomously by "self-prompting." Users can define goals for the AI, and the agent will take autonomous actions to achieve those goals. This includes tasks such as browsing the web, generating original content, and analysing code, all the way through to developing e-commerce businesses.

AutoGPT’s autonomous behaviour is achieved through a feedback loop that involves perceiving the environment, reasoning, taking action, and receiving feedback. Its ability to reflect, learn, and improve its behaviour is a huge step forward for emerging technology and makes GPT models even more useful to businesses than before. Videos of the app writing, debugging and improving code are flying around social media, with the software’s author commenting that he’s working on empowering AutoGPT to debug and improve itself.

AutoGPT is one of several autonomous agents available. As the code for these projects is open source, developers around the world have begun to pick up, adapt and improve the code on an ongoing basis. BabyAGI is another new agent that hit the headlines on release. Described as an AI-powered task management system, the AGI part of its name hints at where we might be heading next.

Agents have long been one of the milestones cited for the achievement of Artificial General Intelligence (AGI). AGI refers to hypothetical machine intelligence that can understand, learn, and apply its knowledge across a wide range of tasks, much like human intelligence. The next theoretical step is sentience. To dive deeper into the philosophical topic of AI and the nature of consciousness, read this insightful new article by friend and colleague Nick Jankel.

The fact that I just linked to a discussion of the nature of consciousness in an article about AI shows how much is at stake here. As business leaders, it’s essential that we navigate this new landscape with foresight and ethics. As these agents become more sophisticated and capable of autonomous decision-making, we have the opportunity to leverage these technologies to drive innovation, improve efficiency, and enhance customer experiences. But it’s equally important to consider the potential impact on the workforce, privacy, and accountability.

The opportunity for us is to learn how to implement AI into our workflow in an environment where intelligent agents work alongside humans to achieve common goals. We need to train our team members to use AI to achieve more impact and effectiveness in their current roles, rather than use AI to remove those roles altogether.

In every industry, AI agents can automate repetitive tasks, analyse large volumes of data, and provide valuable insights to inform our business decisions. The ability of AI agents to learn and adapt over time also opens the door to continuous improvement and optimisation. This is what we’ve been working towards in the software industry for decades.

We truly have the potential to revolutionise industries and redefine the way we work. Yet the power of AI and the pace of its development means that very quickly, it could either transform or decimate the workplace. Which of those two outcomes happens is in many ways up to us. The choices we make now will determine which route we go down as a business community and as a planet.

Conversational, generative and autonomous AI is about to permeate every aspect of our digital worlds, and the tools made available in the next few months will continue to astonish us. We’ll have our own agents working for us while we get on with other things before we know it - most likely before the end of the year. We need to ensure that the ‘other things we get on with are those meaningful things we were born to do - the human, conscious tasks that are our natural talents, our superpowers.

It’s essential now as business leaders to keep abreast of what’s possible and available. The advent of the personal computer levelled the playing field for business. The internet did it again. Then there was mobile. AI, in this new form, looks to change the business landscape in as dramatic a way as those other milestones did. Now’s the time to get stuck in and start seeing how AI-embedded software can advance your business, your industry, and your world.

Blink, and the whole landscape could be different before our eyes. Stay aware, and ensure that you’re part of the change and not a victim of it. Whether AI agents will ever achieve sentience or consciousness remains an open question. Our journey of AI integration promises to be a fascinating and fulfilling one, if we tread boldly, with intelligence and compassion.

Paul McGillivray

Get in touch

If you'd like to talk to me about a project or an idea, get in touch, I'd love to hear from you.

Lets make a difference