In the midst of a heated conversation about AI regulation, leading AI companies are making modifications to their existing models and exploring new ones. It has been reported that OpenAI CEO Sam Altman has proposed that their most powerful language model, GPT-5, could have a hybrid AI model, combining connectionist and symbolic capabilities. This merged approach is what OpenAI is implementing with its new plug-ins for ChatGPT and the Wolfram|Alpha plug-in.
OpenAI, founded by tech influencers, Sam Altman and Greg Brockman, is an AI research firm focusing on developing “Artificial General Intelligence” (AGI) which is a form of AI that can reason and complete tasks similar to humans. OpenAI has worked on products such as Autopilot, OpenAI Gym and Universe, which are meant to reduce the cost of AI engineering. OpenAI’s main open-source project, GPT-3, is seen as breakthrough in the natural language processing (NLP) field and has attracted significant attention.
The Future of Life Institute, an AI think tank, is at the center of the conversation about AI regulation and risk mitigation. Several leading individuals, including Apple co-founder Steve Wozniak, Twitter CEO Elon Musk, and authors Yuval Noah Harari and Yoshua Bengio have signed the institute’s letter calling for a pause on developing more powerful AI than GPT-4 to give the regulatory and safety concerns time to be addressed. Opponents argue that the pause would serve no purpose and the capabilities of current LLMs are already enough for existing AI applications.
Nevertheless, OpenAI may have taken this suggestion into consideration when developing GPT-5. Mr Altman has stated that the Transformer Network technology behind GPT-4 and ChatGPT may have reached its peak, making it unlikely that GPT-5 would incorporate a bigger model. If a hybrid AI approach is employed, it could potentially boost the current capabilities of GPT-4 and GPT-5, while also providing better explainability and accuracy.
DeepMind’s AlphaGo is a prime example of combining connectionist and symbolic AI; it employs a neural network to learn how to play Go while also using symbolic AI to understand the game’s rules. Hybrid AI, however, is still in its early stages and various complications must be met. Nevertheless, combining these two technologies and developments such as OpenAI’s plug-ins can be a huge step forward towards a more reliable and powerful AI.
