What is AI and what can it do?
The short answer is – ‘‘almost anything’’
AI has finally moved beyond its initial ‘‘hype stage’’ and is now providing a range of real-life applications. It possesses the ability to do almost anything using digital text, images, or video. From summarizing books, writing poetry, coding websites, creating exercise plans, winning art competitions, providing medical diagnoses, recommending movies and TV shows based on your viewing history and ratings, gathering visual data in real-time to produce a 3D image that identifies the road, and designing parts of a spaceship to even predicting every previously unknown protein structure, AI does it all.
The image produced with the help of artificial intelligence won 1st place at the Colorado State Fair
Source: CNN
These capabilities are already catalyzing an AI-driven transformation across various jobs, including proofreading, graphic design, sales discovery, programming, and more. The World Economic Forum anticipates that in the next 5 years, AI will transform the core skills of 44% of workers. For example, over the last year, Microsoft’s Copilot AI software has been used to write over a billion lines of code, accelerating task completion for programmers by 55% in the process.
Multiple investment banks, such as Morgan Stanley, have even built AI research assistants for their brokers and analysts. Moreover, individuals can now buy ETFs like the QRAFT AI-Enhanced US Large Cap ETF, which uses AI for stock selection and portfolio rebalancing.
On top of all this, AI will also directly replace many routine jobs such as bookkeeping, data entry, and even certain professional analysts.
I do not think we'll see mass unemployment. But I do think we'll see mass disruption.
— Erik Brynjolfsson, Professor of Economics at the Stanford Graduate School of Business
The rise of personalized chatbots capable of providing tailored information, completing tasks, or even serving as a digital companion during tough times is also evident. In the future, it's conceivable that everyone will have an AI mentor, friend, and assistant.
Everything mentioned is more than amazing. That said, AI is doing so much more, including helping businesses that operate online to overcome strict regulations on gathering personal information for advertising purposes. Through seemingly mundane conversations, AI can actually guess a lot of personal information. For instance, Meta is using its Lattice system to help overcome the setback of Apple’s decision to make iPhone users explicitly opt in to data tracking by the company.
AI is still in its early stages, and as was the case in the first days of the internet, we cannot predict all of the future AI solutions that will become part of our everyday lives. The real revolution will begin once people start leveraging AI algorithms to create new ones – and this is already around the corner with OpenAI’s GPT store.
The advance of technology is based on making it fit in so that you don't really even notice it, so it's part of everyday life.
— Bill Gates
Defining the game – where is it heading?
Stage 1: The origins
AI first captured the public’s imagination way back in the 1950s, when mathematicians started talking about thinking machines. The founders of artificial intelligence believed that learning algorithms would lead to machine intelligence. Yet, the progress proved much more difficult than expected and the hype failed, leaving AI developments closer in resemblance to The Flintstones series rather than The Jetsons.
It was the case until the late 1990s, with the emergence of “machine learning,” that interest in AI was rekindled. In particular, interest was piqued in 1997 when IBM’s Big Blue famously defeated World Chess Champion Garry Kasparov.
Machine learning uses algorithms and probability to identify patterns in data to answer questions and solve problems. The more data it “trains” on, the better it gets at providing accurate answers and solutions.
In 2006, machine learning became “deep learning” when Geoffrey Hinton demonstrated that computers could learn faster by using Artificial Neural Networks – dense layers of algorithms. Having said that, the current AI revolution really began when deep learning systems started using Nvidia GPU chips in 2007 and corporate investment significantly increased in 2010.
Stage 2: The 4th industrial revolution
Since the 2000s, progress has been extraordinary, particularly in the areas of test and speech recognition, i.e. Natural Language Processing (NLP) and image recognition (see chart below).
Language and image recognition capabilities of AI systems since 2000
Source: Our World in Data
Today, the term AI is predominantly used to refer to generative AI models that have been developed with deep learning algorithms. This includes Large Language Models (LLMs) that can read, write, hear, listen, and speak (NLP) and hybrid models that can see and create images and videos (i.e., Image Processing).
How do machine and deep learning alongside language processing form AI
Source: SentiSum
The most widely-known example of an AI large language model (LLM) implementation today is OpenAI’s ChatGPT, which was publicly released on November 30, 2022. Leveraging contextual clues within text, ChatGPT can understand a question or command and provide intelligent and sometimes original responses based on the data the algorithm was trained on.
While GPT-1 and GPT-2 were trained on 117 million and 1.5 billion parameters, respectively, showcasing rudimentary text generation capabilities, the subsequent iterations, GPT-3 and GPT-4, leverage 175 billion and 1.5 trillion parameters, respectively. This, in turn, enables them to pass medical and legal entrance exams and perform better than certain experts in comprehension and writing tasks.
The evolution of parameters used to train OpenAI’s GPT models
Source: OpenAI
The success of ChatGPT has forced OpenAI’s competitors to fast-track the development of their own LLMs. Indeed, Google has introduced Bard and Gemini, Facebook has unveiled LLaMA, and Anthropic has launched Claude. Over the next few years, the major “foundational” LLMs will vie for dominance as the primary platform or “operating system” for AI. Now, we're already witnessing smaller models built from the large, marking a significant trend in the field.
It’s been two months since we announced GPTs, and users have already created over 3 million custom versions of ChatGPT.
– OpenAI