As the AI hype phase wanes, tech companies face pressure to deliver growth despite heavy CAPEX investments in AI. There is an urgent need for these companies to deliver on grand promises of big leaps from AI outside the currently available text and image generating models.
The ultimate goal in this AI race is to achieve Artificial General Intelligence (AGI), where an AI model will have intelligence beyond human levels.
Industry leader OpenAI recently came up with a 5-level classification system for AI models similar to the one used to evaluate self-driving cars.
Level 1: Chatbots with natural conversation abilities, at which stage all the current existing generative models such as OpenAI’s ChatGPT, Anthropic’s Claude, Apple’s Apple Intelligence, and Google’s Gemini can be placed.
Even when a user is interacting with these models and gets a feeling they have some sort of reasoning, they are simply built to predict the next letter, number, or pixel based on the massive amount of data they were trained on.
Level 2: Reasoners with human-level problem-solving capabilities. With advanced reasoning capabilities, a model would be able to analyze, break down, and understand complex problems while also factoring in the human context.
Reuters recently unveiled that OpenAI is working on such a model codenamed Project “Strawberry,” which will be able to access the internet and carry out what the company calls “deep research.”
Level 3: Agents where models would be able to take action independently or with human instructions.
Such models would be the actual virtual assistants we have been aspiring to when Amazon introduced Alexa, Apple with Siri, and Google with Google Assistant. These assistants were only able to execute preprogrammed tasks.
With Level 3 AI, the assistant will be able to independently access the internet, carry out concurrent tasks such as booking a restaurant by checking your calendar, checking your preferences based on your searches or emails, contacting the restaurant, and notifying you once complete.
Level 4: Innovators that can aid in invention. “The kind of Jarvis in Marvel's Iron Man which was able to help Tony Stark find a way to time travel.”
A Level 4 model would be able to find medicine for diseases or design a climate change-resistant bridge with all its technical designs and required materials.
Level 5: Organizations which would be able to do all the work of an organization. Think Gideon running a timeship in DC’s Legends of Tomorrow.
Such a model would be able to run a factory full of robots without any human intervention.
Looking at the progress made so far by all startups and giant tech companies, all of them are still at Level 1, which shows achieving AGI is no easy feat and is going to take quite a number of years if not decades.
Like in the early ages of the internet which peaked with the dot com bust, on the road to AGI, many companies that would have raised millions if not billions of dollars will go down mainly due to the high cost of building AI models and a lack of compelling product-market fit for their tech.
All in all, humanity will be completely changed once AGI is achieved, but one has to hope this milestone does not come at a cost with AI doing more harm than good.