What Is an "AI Company"?
What is an “AI Company”? Many people frequently use that term, yet none ever seem to define what exactly the phrase means.
(public companies mentioned: AMZN, GOOGL, IBM, META, MSFT, NVDA, ORCL)
I think of “artificial intelligence” or “AI” as any form of advanced automation. Robotic Process Automation (“RPA”) qualifies. Predictive Analytics qualifies, especially when machine learning (“ML”) serves as an underpinning. You may be familiar with Netflix’s movie recommendation engine. That is a predictive analytic-based tool powered by machine learning offered via Amazon Web Services (“AWS’). This type of ML-based capability (rooted in Statistics), is pervasive throughout the economy and is deployed across a variety of consumer and commercial services covering every industry.
Underwriters build predictive models to predict risk and outcomes, including weather-related risk, risk of theft and lifespan to name several.
Entertainment companies build predictive models to forecast how content releases may perform at the box office and over streaming services.
Retailers use predictive models to forecast in-store and online sales.
Lenders use predictive models to assess risk around credit worthiness.
All sorts of companies use predictive models to predict customer churn.
Quant shops use predictive models to predict near-term share price movements.
Banks use predictive models to manage risk for a variety of financial instruments.
Technology companies use predictive analytics to monitor and predict user behavior as well as to engage with users (think of Facebook, Instagram and TikTok as examples of companies who tailor content in real-time in order to maximize user engagement). Yet other technology companies use predictive analytics to inform online ad campaigns.
I could go on, the point being that ML-based predictive models have been deployed for years. Therefore, which is the “AI Company”? The company that created the machine learning capability? The company that built and deployed the ML-powered predictive model? The company that leverages this capability to measure performance around a product or service?
Amazon, Google and Microsoft have offered core AI services such as machine learning via their respective cloud server businesses (AWS, GCP and Azure) essentially since they rolled out those cloud server businesses years ago (notice I refer to them as “cloud server” and not “cloud services” as the initial mission for AWS, GCP and Azure was to enable customer companies to rent server capacity with a credit card over the public Internet - i.e. the “cloud”). Oracle also competes here as does IBM.
There are other forms of artificial intelligence such as deep learning, image recognition and natural language processing (“NLP”), all of which have been widely deployed for years across a variety of consumer and commercial use cases. I think of Google as having done more work in these areas than most other companies, as not only does Google offer these advanced automation capabilities inside of GCP, they are embedded throughout Google’s product family from Google Photos to Live Caption to Google Translate to Lens to Maps to Gmail to Google’s Productivity apps and more.
The new wrinkle to AI is the Large Language Models (LLMs) and multi-modal models that are being developed by companies such as OpenAI, Google, Anthropic, Cohere, Mistral, Meta Platforms and to a lesser degree by Microsoft and AWS (both of which are ramping up their proprietary efforts). LLMs are essentially NLP models on steroids. NLP models were typically trained on a particular domain or data set(s), whereas LLMs are trained across domains (the public Internet for example).
Cross-domain natural language capability isn’t so much a novel idea as much as it was OpenAI’s Sam Altman being willing to spend a ton of other peoples’ money to pursue an endeavor without having a product in mind.
Multi-modal models are simply models that marry natural language capability with image recognition as an example.
This new AI flavor of LLMs and multimodal models has been labeled “Generative AI” as the models are capable of generating natural language responses, audio and visual content in response to users’ natural language queries. Generative AI models are not predictive the way in which traditional predictive analytic models are.
Chip companies such as NVIDIA supply the silicon used to power commercial clouds as well as Gen AI models. I suppose it is only fair to give them credit for supplying a key component to the AI value chain.