CEO Tim Cook gave a rare, if guarded, glimpse into Apple’s walled garden during the Q&A portion of a recent earnings call when asked his thoughts on generative artificial intelligence (AI) and where he “sees it going.”
Cook refrained from revealing Apple’s plans, stating upfront, “We don’t comment on product roadmaps.” However, he did intimate that the company was interested in the space:
“I do think it’s very important to be deliberate and thoughtful in how you approach these things. And there’s a number of issues that need to be sorted. … But the potential is certainly very interesting.”
The CEO later added the company views “AI as huge” and would “continue weaving it in our products on a very thoughtful basis.”
Cook’s comments on taking a “deliberate and thoughtful” approach could explain the company’s absence in the generative AI space. However, there are some indications that Apple is conducting its own research into related models.
A research paper scheduled to be published at the Interaction Design and Children conference this June details a novel system for combating bias in the development of machine learning datasets.
Bias — the tendency for an AI model to make unfair or inaccurate predictions based on incorrect or incomplete data — is oft-cited as one of the most pressing concerns for the safe and ethical development of generative AI models.
So glad OpenAI is keeping its bias in check. pic.twitter.com/y4a7FUochR
— Brooklin Nash (@realBrookNash) April 27, 2023
The paper, which can currently be read in preprint, details a system by which multiple users would contribute to developing an AI system’s dataset with equal input.
Status quo generative AI development doesn’t add in human feedback until later stages, when models have typically already gained training bias.
The new Apple research integrates human feedback at the very early stages of model development in order to essentially democratize the data selection process. The result, according to the researchers, is a system that employs a “hands-on, collaborative approach to introducing strategies for creating balanced datasets.”
It bears mention that this research study was designed as an educational paradigm to encourage novice interest in machine learning development.
It could prove difficult to scale the techniques described in the paper for use in training large language models (LLMs) such as ChatGPT and Google Bard. However, the research demonstrates an alternative approach to combating bias.
Ultimately, the creation of an LLM without unwanted bias could represent a landmark moment on the path to developing human-level AI systems.
Such systems stand to disrupt every aspect of the technology sector, especially the worlds of fintech, cryptocurrency trading and blockchain. Unbiased stock and crypto trading bots capable of human-level reasoning, for example, could shake up the global financial market by democratizing high-level trading knowledge.
Furthermore, demonstrating an unbiased LLM could go a long way toward satisfying government safety and ethical concerns for the generative AI industry.
This is especially noteworthy for Apple, as any generative AI product it develops or chooses to support would stand to benefit from the iPhone’s integrated AI chipset and its 1.5 billion user footprint.