Working for a Microsoft Partner, and having gone to university for a philosophy degree, I thought I’d take a bit of time today and discuss a bit of an interest of mine (as well as an interest of tech geeks everywhere…) and that’s the fascinating and rapidly evolving field of Artificial Intelligence (AI) and its journey towards Artificial General Intelligence (AGI).
I know tech and philosophy seem like an unlikely pairing but in this instance, the two couldn’t be more inextricably linked.
Thanks to Chat GPT’s very public launch last year, AI has rapidly entered the public consciousness, with the potential to be the biggest transformative force since, well, the wheel; revolutionising industries and redefining the boundaries of what machines can accomplish in the span of just months or even weeks and days… although what many don’t realise is that it’s actually been around, influencing our lives, for a while now already.
AI, as we know it today, covers an enormous range of technologies and methodologies, all aimed at creating systems capable of performing tasks that would normally require human intelligence (or at least a human).
From voice assistants like Siri and Alexa to recommendation algorithms powering personalised content on streaming platforms, right through to Chat GPT, AI is already well and truly integrated into our daily lives.
But, as impressive as those apps may be, they’re still known for being what’s called narrow or weak AI, designed to excel at specific tasks only, within predefined parameters.
Artificial General Intelligence (AGI), on the other hand, would represent a hyper leap forward in AI capabilities, as its aim is to develop machines that will possess the ability to understand, learn, and apply knowledge across a wide range of functions, mirroring the broad cognitive capabilities of human intelligence.
Those responsible for creating AGI systems are trying to create machines capable of reasoning, problem-solving, abstract thinking, and adaptability, without the need for the pre-set parameters narrow AI requires to function.
Just imagine a system that can not only recognise objects in images, but can also understand the context of a query, generate creative solutions, and learn from its experience in a manner similar to human cognition.
Achieving a true AGI could unlock immense potential, revolutionise industries, advance scientific research and even reshape the future of humanity.
It’s important to note however, that whilst AI’s made remarkable strides in recent years, the journey towards AGI is still very much ongoing.
Written by – Thomas Cunningham
Development Practice Lead, FormusPro
Understanding AI: The Basics
If we’re going to be talking about the evolution of AI into AGI, it's probably best to establish a solid foundation of AI principles; what are they and what are they used for:
Something called Machine Learning (ML) is what lies at the core of all modern AI, enabling systems to learn patterns and make predictions without being explicitly programmed.
ML algorithms analyse vast swathes of data to identify underlying patterns and make informed decisions around them… within the scope of their programming.
One obvious example of that could be image recognition, in which convolutional neural networks have revolutionised tasks such as facial recognition, object detection, and autonomous vehicle perception.
Neural Networks are also a key component of AI, inspired by the structure of the human brain.
Deep Learning, a subfield of ML, utilises neural networks with multiple layers to process complex information hierarchically.
This approach has yielded impressive results in natural language processing, speech recognition, and even defeating human champions in complex games like Chess and Go.
In fact, AlphaGo was a program developed by DeepMind back in 2016, that defeated the world champion Go player, Lee Sedol, in a groundbreaking milestone for AI.
Through a combination of deep neural networks and reinforcement learning, AlphaGo showcased the power of AI by mastering a game that many thought required intuition and strategic thinking just to play, let alone master, and would always be beyond the reach of a machine.
AI's impact isn’t just limited to gaming though.
In healthcare, AI clearly demonstrated its potential to improve diagnosis accuracy and treatment outcomes when researchers at Stanford University developed a model that could diagnose skin cancer with an accuracy comparable to trained and experienced dermatologists.
Finance is another sector that’s witnessed huge transformations thanks to AI.
High-frequency trading algorithms employ AI techniques to analyse vast volumes of market data and make real-time trading decisions. Those algorithms are now easily capable of exploiting market patterns to execute trades at speeds far beyond human capabilities, improving both efficiency and profitability.
AI expert and computer scientist Andrew Ng often says, "AI is the new electricity."
Just as electricity transformed numerous industries, AI is poised to reshape how we live, work, and interact… but can AGI do the same?
Getting From AI To AGI
The path from AI to AGI is going to be a long and ongoing, but exciting journey.
Over the last few years, researchers and engineers have made significant strides in advancing AI systems, bridging the gap between narrow AI and the broader capabilities that will be required for AGI, but there’s a long way still to go.
One crucial aspect of this evolution has been the refinement and advancement of machine learning algorithms.
Traditional machine learning approaches relied heavily on labelled data, where human experts provided explicit annotations for the training process.
However, advancements in unsupervised learning techniques, such as Generative Adversarial Networks (GANs) and self-supervised learning, have massively reduced the dependency on labelled data and allowed AI systems to learn from raw, unannotated data.
Deep neural networks have played a pivotal role in this journey.
The introduction of deep learning architectures, with multiple layers of interconnected neurons, enabled the development of complex models, capable of extracting high-level features from raw data. We can easily see the success of deep learning in various domains, such as computer vision, natural language processing, and speech recognition.
Transfer learning, another huge leap forward, has allowed models trained on one task to leverage their acquired knowledge and adapt to new, related tasks more efficiently. This capability tries to mirror human learning, where prior experiences influence the learning process in new situations.
For example, OpenAI's GPT models, initially trained on a massive corpus of text, can then be fine-tuned for specific tasks like language translation or document summarisation.
Reinforcement learning, a branch of AI focused on learning through interaction with an environment, has also contributed to the evolution towards AGI.
By training AI agents to maximise rewards, whilst navigating complex environments, reinforcement learning has enabled breakthroughs in game playing, robotics, and autonomous systems.
One remarkable example is the reinforcement learning algorithm used by DeepMind to teach an AI agent to play complex video games, surpassing human-level performance in games like Dota 2.
Despite these advancements however, AGI remains an ambitious and nebulous goal with significant challenges.
AI systems still lack the ability to generalise knowledge beyond their training data, struggle with common-sense reasoning (although to be fair to AGI, many humans I know still struggle with this) and face many ethical dilemmas.
For AGI to truly evolve, it’ll require huge steps forward in cognitive architectures that can integrate knowledge from diverse domains and exhibit flexible and creative problem-solving abilities.
Professor Stuart Russell, a well-known AI researcher and author, emphasises the importance of aligning AGI's objectives with human values.
He suggests that AGI systems mustn’t merely optimise for a predefined goal but should consider human preferences and values to ensure they benefit humanity as a whole.
Why Achieving True AGI Is So Hard
The pursuit of AGI has always created significant challenges for researchers and engineers to overcome, most of which have never applied to AI.
Whilst all our advancements in AI may have brought us closer to AGI, several, rather large hurdles still remain on the path to achieving true, human-level intelligence in machines (I’ll get to Skynet in a bit!).
As I’ve already circled around, one of the biggest challenges will lie in enabling AI systems to generalise knowledge beyond their training data. Whilst current AI models excel at specific tasks within limited domains, they often struggle to transfer their learnings to new and unfamiliar situations.
For example, an image recognition AI trained on photos of cats may fail to identify a cartoon illustration of a cat or a real-life image taken with an unusual perspective.
Generalisation will require AI systems to understand abstract concepts, handle uncertainty, and apply knowledge in novel and unusual contexts with no human input to provide it.
Common-sense reasoning is another of the biggest challenges yet to be solved.
Humans possess a vast repository of common-sense knowledge acquired through everyday experiences. But encoding this tacit knowledge into AI systems remains elusive.
A human could easily understand that "it’s impossible to fit an elephant into a shoebox," based on their common-sense understanding of the physical world. Teaching AI systems to reason based on implicit knowledge and grasp contextual nuances remains an ongoing research area.
One of the largest areas to worry developers of AGI currently though is the ethical concerns surrounding its potential growth.
As AGI evolves, it’s vital to ensure that the technology is developed responsibly and aligns with human morale’s.
An ethical framework should address issues such as privacy, security, fairness, transparency, and accountability.
There’s been a lot of talk lately around AI systems that rely on biased training data that perpetuate societal biases or discriminate against certain groups. Researchers and policymakers are actively working to establish guidelines and regulations to navigate these ethical challenges but there’s still work to be done.
Another challenge will be in defining the control and decision-making mechanisms for AGI.
As these systems become increasingly intelligent and autonomous, it’ll be crucial to design mechanisms that align with human values and preferences, ensuring any AGI acts in a way that is safe, beneficial, and in accordance with human values, as almost all experts agree on the need avoid potential risks and unintended consequences.
On the hardware side of things. AGI development will also demand immense computational power and resources.
Training complex models with billions of parameters requires significant computational infrastructure and energy consumption. The development of efficient algorithms and hardware architectures is essential to overcome these challenges and make AGI more feasible and accessible.
AI pioneer and computer scientist, Professor Yoshua Bengio, has repeatedly emphasised the importance of long-term safety and the need for research on robust and explainable AGI.
He suggests that understanding and controlling AGI's decision-making processes are crucial for its responsible deployment. Addressing these challenges will require collaboration amongst researchers, policymakers, ethicists, and even society at large.
Organisations like OpenAI have embraced cooperative approaches, seeking to ensure AGI benefits all of humanity and avoids harmful applications but it’s still too early to tell how successful this will be.
The Future of AGI: Implications, Possibilities And Scary Outcomes!
As the journey towards AGI continues, it's become crucial to explore the implications and possibilities that could (and do) lie ahead.
The development of AGI has the potential to reshape our society, industries, and the very fabric of human existence. In fact, one of the most profound implications of AGI is in its transformative effect on industries and the economy.
AGI-powered automation will revolutionise various sectors, from manufacturing and logistics right through to healthcare and agriculture.
Repetitive and labour-intensive tasks can be efficiently performed by intelligent machines, liberating humans to focus on creative, strategic, and high-level decision-making roles.
The shift brought about by AGI has the potential to redefine job roles, enhance productivity, and unlock new avenues of economic growth but if not managed carefully could leave millions of minimum wage workers unemployed.
Moreover, AGI's impact on scientific research and innovation will be immeasurable.
AGI systems can assist scientists in accelerating discoveries and making breakthroughs in complex domains such as drug discovery, materials science, and climate modelling. The ability to analyse vast amounts of data, simulate complex systems, and generate novel hypotheses can lead to advancements that were previously unimaginable.
However, all these advances raise huge ethical considerations.
The potential risks associated with AGI, such as unintended consequences, misuse, and control-related issues, let alone rise of the machine type scenarios, must be carefully addressed. Collaboration among researchers, policymakers, and ethicists is essential to establish robust frameworks and guidelines that mitigate risks and foster responsible AGI development.
It’s crucial to ensure that AGI benefits are distributed equitably and that access to AGI technologies isn’t limited to a privileged few whilst leaving millions dispossessed. Efforts must be made to bridge the digital divide, provide education and training opportunities, and address potential socio-economic disparities that will arise from large scale AGI adoption.
I’m not saying that the future possibilities of AGI aren’t awe-inspiring.
They’ll aid in tackling complex global challenges, such as climate change, resource management, and space exploration. The very systems themselves could even collaborate with humans to find innovative solutions, enhance our understanding of the universe, and push the boundaries of scientific exploration but we also need to answer the philosophical and existential questions about the nature of intelligence, consciousness, and the human experience first, which is a problem as the evolution of AGI keeps rushing forward faster and faster.
Ethical Considerations And Responsible AGI Development
As the field of AGI progresses, it’ll become paramount to lockdown these ethical considerations and ensure responsible development practices.
It has the potential to bring about enormous transformative changes in society, but we cannot forget that its deployment has to align with our values, respect our individual rights and safeguard against potential risks (especially around privacy).
That’s why one of the key ethical considerations is the potential impact AGI could have on privacy and data protection.
AGI systems will (and do) rely on vast amounts of data to learn and make informed decisions. Ensuring that an individuals' personal information is handled with utmost care and that suitable privacy safeguards are in place, will be crucial.
Striking the right balance between leveraging data for AGI advancements and respecting privacy rights is something that isn’t being discussed enough alongside this technology.
Transparency and explainability are also vital ethical concerns.
As AGI systems become more complex and capable, understanding their decision-making processes becomes increasingly important.
Researchers and developers must strive to create AGI systems that aren’t locked away black boxes, but instead can provide clear explanations for their actions and reasoning.
This transparency will be crucial for users to trust and comprehend the tech so we can all identify potential biases or unintended consequences.
One case study that highlights the significance of transparency is the COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) algorithm used in the US criminal justice system.
The algorithm, aimed at predicting the likelihood of a defendant committing a future crime, came under a lot of scrutiny due to concerns about its biased outcomes.
By analysing the underlying mechanisms and data used by such algorithms, we can better understand the potential ethical pitfalls and work towards fair and unbiased AGI systems.
Fairness and inclusivity should also be a top concern for anyone involved in the development of AGI.
AI systems can’t do much on their own if there’s underlying biases present in the data they’re trained on.
If said data contains biases, such as gender or racial bias for example, the resulting AGI systems would inadvertently perpetuate and amplify those biases in their decision-making processes. Proactive efforts need to be made to address and mitigate such biases to ensure fairness and equal treatment for all individuals.
Another crucial ethical consideration is the potential impact of AGI on employment and socio-economic disparities. As AGI technologies automate various tasks, there’s the very real possibility of job displacement on a record scale for certain sectors.
Governments need to start proactively addressing the societal consequences of these shifts, ensuring that measures are in place to support affected workers through retraining programs, reskilling initiatives, and social safety nets such as universal basic income policies.
No one government can do this however.
Responsible AGI development almost demands international collaboration and cooperation. The impact of AGI transcends national boundaries, and global frameworks and guidelines need to be established soon to ensure ethical standards are upheld consistently.
International organisations, governments, and research institutions have to work together to address these ethical challenges to foster responsible development.
To promote responsible AGI development, organisations like OpenAI have published guidelines and principles.
OpenAI's Charter emphasises a commitment to using any influence they obtain over AGI to ensure it’s used for the benefit of all and avoids uses that could harm humanity or concentrate power unfairly.
In summary, the development of Artificial General Intelligence (AGI) holds tremendous potential to transform our society and shape the future of humanity but as we embark on this journey, it will be crucial to consider the ethical implications, responsible development practices, and the well-being of individuals and society as a whole.
The future of AGI is not a solitary endeavour but a collective one that requires the collaboration and expertise of individuals from incredibly diverse fields.
Together, we can shape a future where AGI serves as a powerful tool to enhance our lives, push the boundaries of knowledge, create a more inclusive and sustainable world and doesn’t leave us obeying our machine overlords.