Will Artificial Intelligence take over Humanity?
While the above title seemed preposterous and I would normally discard being associated with the field of Artificial Intelligence [AI], however, when Geoffrey Hinton, godfather of Artificial Intelligence, resigned from Google to warn the world about dangers of AI outperforming humans and eventually taking over [1]; it warranted a second look. The key reason of his decision was the realization that AI outperforming humans is much closer to his initial estimate of 30–50 years as some of the large Generative AI have already achieved human like intelligence (in most cases).
Geoffrey Hinton[2] is considered as Godfather of Artificial Intelligence due to his work on back propagation which enabled the Artificial Neural Network to learn from data and on deep learning which have revolutionized object classification and speech recognition.
Let us first understand what different terms like Artificial Intelligence, Artificial Neural Networks mean.[3][4]
Artificial Intelligence is like a discipline concerned with development of systems that mimic human intelligence. These systems can also have rule-based engine at their core. However, it is not the rule based artificial intelligent system that are making all the news. A subset of Artificial Intelligence is ‘Machine Learning’. This is associated with having a model at the core which is fed with the data and the model learns the pattern on the data. This model can now make predictions when fed with new data. The erstwhile machine learning models were mostly having a core algorithm be it a linear or a polynomial line which is fed with the data to find its parameters (e.g. slope and intercept in case of linear model which mimics a straight line) or a tree structure or statistical model predicting value based on its nearest neighbours etc. These models met with some success but were soon identified to be limiting in terms of identifying complex patterns like object detection or speech recognition. This led to another subbranch of machine learning called Deep Learning. These had Artificial Neural Networks (ANN) at its core. An Artificial Neural network mimics human brain and can be trained with internet scale of data to understand complex patterns required in vision, speech and language.
Key Characteristics of Artificial Neural Network (ANN) -
- Inspired by biological neural network present in central nervous system and brain.
- Human brain is densely interconnected network of approx. 10 ^11 neuron interconnected with other 10 ^4 neuron.
- Each neuron is capable of conditioning [increase of decrease the strength of signal] and cause excitation or inhibition of subsequent neuron.
- Artificial Neuron Network simulates biological neural network as set of connected multilayered input/output layer. The connection between neuron has weight associated with it.
- As the input signal gets progressed thru the ANN, it gets conditioned by the weights and activation function of each neuron and finally output is produced.
- The training of the artificial neural network is essentially adjustment of weights [and selection of activation function] to be able to predict the correct output.
All the recent complicated models in the area of Vision, Speech, Language (e.g. Chat GPT) have ANN at their core. There are various variations derived on top of ANN like Convolution Neural Network, Recurrent Neural Network, Encoder-Decoder, Generative-Adversarial Network (GAN), Transformers etc. Recently a subbranch of Deep Learning called Generative AI which is based on Transformer models have generated lot of interested due to their ability to create and train large models consisting of trillions of parameters which started to mimic human like intelligence. These models have been reported to understand jokes and even respond like a highly intelligent individual. In fact, one of the google engineer Blake Lemoine felt as if the chatbot based on google LaMDA model was really a sentient having a soul based on his interaction with the chatbot. [5]
Even if we keep sentient kind of dominance aside, the AI can be used to develop sophisticated programs which can easily interfere, and influence elections given the ease with which it can produce disinformation. AI generated images and videos are already creating ripples. On 22 May 2023, a fake AI generated image of smoke behind pentagon triggered a stock sell off in the US markets. Due to the fake AI image S&P 500 fell down by 0.3%. [6][7]
In Apr2023, around 30,000 AI researchers and other academics signed a letter calling for pause on AI research so that the risks of society are better understood, and we find a way to create mature, ethical AI by developing the guardrails. [You can read about applying Ethical AI principles here] [8]. The pause, however, could not be implemented as it is very difficult to regulate it across borders and no one wants to give up on first mover advantage given the benefits AI promises.
Let us tackle the question of how far are we from a machine taking over if we continue the same path?
Why Humans Rule the Earth?
First let’s try to understand what makes Humans rule the earth and not by some other animals. There are animals which are more powerful and agile than humans like Cheetah, Eagle, Tiger or even have a bigger brain than humans like Elephants or Whales. In the book ‘Seven and Half Lessons about the brain’ [9] — the author renowned neuroscientist Lisa Feldman Barrett has provided a brief history of how human brains evolved backed by scientific research. One of the key things that differentiates human brains from other animals is the ability to learn. While most of the animals’ offspring learn about their tasks before being born or in other words their neural networks are hard wired; humans learn majority of its tasks after being born with their ability to focus, learn and unlearn depending upon the environment human babies are put it. This ability to learn (or unlearn) depending upon the environment gives the edge to humans over other animals and eventually makes us the most powerful species to rule the earth.
What is so special about Artificial Neural Networks?
Now if we take the large Neural Network models, these models mimic human brain structure and with recent advance in form of transformer-based models, we can train large models which has more neurons than human brains with the internet scale data. These points are very important since it suggest that we can create artificial brains that are larger than the size of humans and train (and untrain them) them with all the data of the world. This perhaps is best of both world as we can now have brains the size of Elephant (or a Whale) with the ability to learn and unlearn depending upon the experience like humans. Any species having this powerful brain is ought to rule the earth.
Dependency on Humans
It is true that current AI machines need sophisticated form of energy (electrical) for them to work but we as humans also cannot extract the energy from the food that we eat without the help of billions of bacteria that operate in our alimentary canal especially in gut in a symbiotic manner. So, it is also possible that humans become so dependent on AI machines for critical tasks (e.g. operating nuclear power plants) so as to keep supplying them with the electricity. While AI systems can use this energy to not only ostensibly perform the so-called critical task assigned by humans, but also in parallel develop another eco-system with other AI systems over network and eventually control humans.
One can still argue that machines cannot reproduce like humans, however, history of colonization tells us that the rulers were always far and fewer than the ruled!!
On the positive side, if we can leverage it rightly — this also can become a tool to prevent alien invasion by more intelligent species. Homo sapiens have been able to control other species which are more powerful and skilled (in some areas), time will tell if it can control more intelligent systems as well. We certainly cannot stop the AI revolution now; however, by imbibing principles of Ethical AI[10], we can certainly create AI systems that are more responsible towards humans. This can certainly keep humans in control of AI and would ascertain that we continue to leverage them for betterment of mankind.
Abhinav Ajmera
Enterprise Architect,
AI, Data and Cloud.
References –
2. https://en.wikipedia.org/wiki/Geoffrey_Hinton
3. https://en.wikipedia.org/wiki/Artificial_intelligence
4. https://en.wikipedia.org/wiki/Artificial_neural_network
5. https://www.npr.org/2022/06/16/1105552435/google-ai-sentient
6. https://www.nytimes.com/2023/05/23/business/ai-picture-stock-market.html
8. https://abhinav77.medium.com/ethical-ai-applying-the-7-principles-2d45f165d20b
9.https://sevenandahalflessons.com/notes/Extended_notes_for_Seven_and_a_Half_Lessons_About_the_Brain
10. https://www.capgemini.com/insights/expert-perspectives/ethical-ai-decoded-in-7-principles/
