How Businesses are Putting Generative AI to Work

Photo by Luca Bravo on Unsplash

Prompt ChatGPT to write a love poem in the style of Shakespeare from the perspective of a Lovecraftian monster. Go on. See what it says. But if you refuse to challenge ChatGPT’s romance novel-writing skills, then perhaps you’d join the rest of its 100 million active monthly users in asking it to summarize a history passage in a late-night study sesh or plan a surprise party for your best friend’s birthday.

Generative AI seems to be everywhere nowadays. ChatGPT, Midjourney and Amazon’s Alexa are just some of the applications of generative AI that we use on a daily basis. It’s understandable: generative AI is fun, basically self-aware from the user’s perspective and seemingly full of possibilities. But for a term that’s thrown around so much, it sure doesn’t take on a clear meaning. Let’s clarify what generative AI is and discuss its exciting applications outside of home and into the workplace.

What is generative AI?

Generative AI refers to deep learning models一a subset of Machine Learning (ML) also known as neural nets一that learns from raw data to generate new, original content which can take the form of text, images or video. However, Generative AI can’t just create anything out of thin air. Rather, it generates the most statistically probable outcomes.

Generative AI’s reliance on the statistically probable is exactly why ChatGPT can be so grossly inaccurate at times in its responses (there’s actually a term for these unjustified responses called “AI hallucinations”) and at complex math. Though Large Language Models (LLMs) such as ChatGPT may appear to understand human language and logically reason, they simply cannot. These models sacrifice accuracy for clarity to such an extent that the user is fooled by their impressive command of language and imagines them to be credible sources when they are not. Deep Fakes illustrate pitfalls of generative AI in a more playful manner, swapping the face of a web influencer on a controversial tirade for that of a respectable politician and releasing all hell on the internet through these viral, yet fake videos.

However, that is not to say that generative AI is a disingenuous tool that will yield more negative social change than positive. Those responsible for developing friendly AI generally seek to eliminate bias when training or building its models and those using AI should always keep a watchful eye for misinformation. Besides, generative AI is just far too convenient and powerful to throw away, which is exactly why it has also attracted the attention of large tech companies in Silicon Valley.

How is generative AI being used in the workplace?

AT&T is one tech company building its own chatbot from OpenAI’s language model systems after noting the widespread use of ChatGPT amongst its developers when debugging code. Out of safety concerns, such as the risk of leaking sensitive information on a publicly available tool, AT&T turned to Azure Open AI Services to create Ask AT&T, which was originally intended to facilitate the coding process for its developers. Soon enough, however, employees working in customer service began using the chatbot to summarize meetings and calls and write repetitive emails.

Alternatively, on the receiving end of customer service, generative AI is personalizing human-chatbot interactions with its ability to speak in and translate dozens of languages and quick, context-appropriate responses. Obviously, there is still room in improvement for chatbot-powered customer service in general with 88% of U.S. consumers preferring interaction with a human representative over a virtual one. Yet, it is only expected for AI to play a more prominent role in customer service in the future, expanding access to an otherwise busy support line and placating thousands of impatient customers.

In terms of the medical field, AI chatbots have already proved promising in the accurate self-diagnosis of diseases. Google’s LLM Med-PaLM 2 was designed to safely answer medical questions and expand access to accurate medical advice to developing countries. Its chatbot has scored over 85% on the US Medical Licensing Examination (USMLE), far above the passing score of 60%. Moreover, when epidemiologist Andrew Beam of Harvard University fed 48 prompts of descriptions of patients’ symptoms to OpenAI’s GPT-3 model, the top three most probable diagnoses for each case contained the right diagnosis 88 percent of the time. Licensed physicians fared only a bit better, producing an accurate diagnosis 96% of the time under the same task. What’s more certain is that these AI Chatbots can guide patients away from less reliable online symptom checkers and ask follow-up questions to clarify potential symptoms. Though the chatbot itself cannot supplement the medical advice of a doctor, it can offer a more “human” approach to self-diagnosis and ultimately produce more accurate results.

Lastly, biotech companies are exploiting generative AI in drug discovery by efficiently hypothesizing new molecule structures for ligands, which target specific proteins and inhibit chemical pathways that can lead to the development of cancer. According to IBM research, it is believed that “there are 10^63 possible drug-like molecules in the universe”. Traditional trial-and-error development methods will not even break through a billionth of those combinations. Worse, 90% of theoretically successful drugs fail when tested on humans and those failures translate to hundreds of thousands of dollars lost by biotech companies in the process. If generative AI models can predict new molecular arrangements or determine the structure of 3D proteins just from the sequence of its nucleotides, then the problem of timing and excessive cost may be circumvented. Now, scientists may choose to manually test a smaller batch of compounds in the lab, then gather feedback from a generative AI model that predicts the next hundred statistically probable candidates.