We have come a long way with AI. Chat-GPT that uses a large language model (LLM) is nothing but an early chapter in the story of AI. Chat-GPT and other AI assistant Chatbots have several applications but they do hallucinate facts and manifest that they lack planning and reasoning skills required to complete high school level math problems. Chat-GPT is a machine learning or deep learning program that works efficiently on a given task. The neural networks involved in it showed emergent properties. It was allowed to learn on the data sets of Amazon reviews. It showed an emergent property of identifying positive and negative sentiments which are very important in the market It opened us possibilities of working on large language Models (LLM) and was opened to millions of people to experience the same. There are limitations to it. It is bad in poetry ( it can write poetry but no beat humans for now) as well as mathematics. The promise of the Chatbot technology is already opening our imagination to the next generation of AI.
Although, we are in the weak AI mode, there are early signs that we are getting ready to bring in a strong general AI ( AGI). We can notice these signs in what happened to Sam Altman, the CEO of OpenAI. Altman was fired by the board of OpenAI citing some vague reasons only to be reinstated immediately because of the revolt of other employees. But this action does not really have a logical explanation and hence, some people have indicated about the issue of being deeper than what meets the eye. It turns about to be exactly the case. OpenAI was working on a new AI which was internally named as Q* (Q star) which was to be a step closer to general AI and which would do several tasks better than human beings. Altman had already predicted that all repetitive human tasks will be performed better by AI and Q* was being prepared to out-perform human beings on many fronts. It is said that the board was afraid that Altman might charge ahead productizing Q* without giving adequate time for safety trials.
How Q* works is only an intelligent guess. Some AI scientist like Jim Fan is saying that Q* is making use of several AI models working together to learn, plan and carry out tasks. It is allied to DeepMind’s AlphaGO that defeated the world champion Go player in 2016 that used several neural networks working together and learnt by playing a millions of games against an older version of itself. Q* says Fan may rely on similar architecture. OpenAI is not alone in taking this team approach of AI systems. DeepMind itself is working on a new AI system called Gemini which CEO Demis Hassabis has suggested that it might make use of a similar approach to that of AlphaGo but with a large language model (LLM) thrown into the mix. Thus, Q* and Gemini appear to be the next generation Chatbots. What we are witnessing is an AI race between top giants in the field. Amazon AWI had announced its own AI Chatbot-Q while Google has presented its Bard. Amazon Chatbot-Q is not intended to general public but is designed to be used by a large work force that needs to access and synthesize large company’s corporate data . AWS cloud store houses and keep safe such data and offers Chatbots that will not leak that data when used by the employees. These new Chatbots are likely to prove to be disruptive technologies in the coming days.
The future of Chatbot technologies is coming faster than we can anticipate. How are we to respond to these developments. We cannot exhibit a phobic response nor can be simply blindly go gaga about it. What we need is prophetic acceptance. Prophetic acceptance is a critical response that studies the technology critically and uses it for the good of humanity. Although there is no consensus among AI scientists on the fact of technologies like Chat-GPT as working with language semantically, it appears that these technologies works with a kind of distance association that enables the program to predict the next word that it is most fitting or close in a given sequence of words. Hence, it appears that it is through computational sequencing that Chat-GPT can correlate words and compose coherent and relevant answers to our questions. Deepfakes indicate that it is not just words that are manipulated by these kinds of models. They show that these kind of models can also manipulate and generate our digital clones that will then do and say what we have never said and did based on the data sets of our life that are already digitally present. This means there are possibilities of abuse of these disruptive technologies. Hence, we need ethics to guide their use. This is why we need prophetic acceptance of these new AI that is all set to change our life in the coming future. With growth of self-supervise learning AI , we are not far from machines mimicing and outsmarting us.