Within a short period of time, businesses, academic institutions, and various creators are finding incredible applications for generative AI, and with each new version introduced, it seems its capacity to handle massive amounts of data increases. But this is where the future of generative AI is less clear. Debates exist regarding whether Small AI versus Big AI is the best direction for ongoing developments. Both have benefits that could be considered, but it’s clear there’s a battle to dominate AI for each. And ultimately, it depends on a variety of factors as to which succeeds.
(Generative AI and copyright law are clashing–read up on the conflict in this Bold story.)
By definition, the distinction between Small AI versus Large AI is one of parameters. Large language models are those that use dozens of billions of parameters to train and guide generative AI models. In contrast, smaller systems may only utilize a billion parameters and accomplish many of the same tasks. Whether larger or smaller platforms are best for the future depends on who you ask. Notably, there is significant debate concerning the subject. But the impact of which approach wins the battle to dominate AI may well be important. This is especially true in terms of whether or not tech giants of today will continue to exert power over the industry. Thus, exploring factors that may influence the outcome between Small AI and Big AI is a worthwhile endeavor.
The Pros and Cons of Big AI
When it comes to the battle to dominate AI in the future, there’s no doubt Big AI can make a strong case. Large language models (LLMs) used to train and operate such platforms like ChatGPT and Bard are sizable. These each use over 100 billion parameters, which enables Big AI to perform some pretty impressive tasks. It is precisely this capacity that support predictions that Big AI may one day match or even exceed human abilities. As their digital brains expand into evermore connections, they may acquire what is known as artificial general intelligence. In the Small AI versus Big AI debate, those who foresee AI’s overall potential expect Big AI to be the obvious choice.
While the potential for Big AI to win the battle to dominate AI is noteworthy, there are some negatives associated with LLMs. For one, they demand a tremendous amount of energy, which will require new advances in energy to keep pace. As more and more parameters are added to AI training and operations, power demands increase. Already, the use of ChatGPT use on a daily basis consumes enough electricity to provide 33,000 homes with electricity. If Big AI was used to support global searches on Google, its energy consumption would be comparable to that required by Ireland as a nation. Plus, storage is also an issue as is construction. Higher amounts of parameters make development more difficult, and LLMs require cloud-based storage. These are not insignificant issues from a practical perspective. In the Small AI versus Big AI outlook, this could give the nod to the former.
The Pros and Cons of Small AI
As expected, Small AI is significantly less involved and complex when compared to Big AI. These platforms tend to train on a billion parameters or less, and they naturally consume less energy. In addition, however, Small AI is still quite adept and can perform many of the tasks that Large AI performs. In fact, some experts suggest that parameters of Big AI models could be reduced by 60% without losing significant performance. Not only does this favor Small AI versus Big AI based on complexity and power. But it also means Small AI would be less expensive to develop and use when compared to LLMs. This explains why some expect Small AI to have an advantage in the battle to dominate AI in the years to come.
Indeed, the smaller capacity of Small AI means that it will not be able to handle the larger AI pursuits as described previously. Likewise, it also means that scaling larger use of AI through various hardware and software is less likely. At the same time, however, this permits a more targeted use of generative AI for personalized endeavors and smaller tasks. Not only could this align well with other individualistic trends, but it would also encourage more open sourcing. Plus, the use of Small AI applications on a specific device would have fewer cybersecurity concerns when compared to Large AI systems. Notably, there are a lot of pluses for Small AI in the battle to dominate AI moving forward. Despite this, the Small AI versus Big AI debate continues because of the major players involved.
The Politics of AI
The early leaders in generative AI thus far look to be the tech giants of old. Microsoft has significant stock in Open AI, and Google and Meta are actively pursuing Big AI. But that doesn’t mean they’re not hedging their bets since the outcome of Small AI versus Big AI is unknown. For example, Microsoft has already announced two Small AI projects, Phi-1.5 and Phi-2. The ease of development and lower costs associated with these projects explain why it’s easy for these companies to pursue both. Just in case Big AI loses the battle to dominate AI of the future, they want to be prepared. Thus, it’s not too surprising such mutual pursuits are ongoing.
(Personalized AIs are here–what could go wrong? Read this Bold story and find out.)
When all is said and done, however, it’s likely Google, Meta, Microsoft and even Apple hope Big AI wins out. Why? Because they stand to gain a great deal. Only these types of companies capable of investing tremendous amounts of money can play in the Big AI game. Should Big AI succeed in the battle to dominate AI, these companies’ tech dominance will likely persist for decades. At the same time, should regulatory oversight develop over AI, these same companies will be the ones able to afford compliance. While this perhaps sounds disheartening, it does mean the Small AI versus Big AI war is settled. Ultimately, consumers will determine whether they prefer personalization and miniaturization or highly advanced AI.
