TECHSPECTATIONS 2023

Google, Amazon and Adobe honchos speak of AI's dark side that gives them sleepless nights

Here is one thing about the biggest fad of our times, Artificial Intelligence (AI), that keeps top executives at Google, Amazon and Adobe awake at night. Misinformation.

“If there is anything that keeps me up at night, it is the scare of misinformation,” said Subram Natarajan, Director of Customer Engineering, Google Cloud. He was speaking during the panel discussion on 'Artificial Intelligence & Business Intelligence' organised as part of Manoramaonline Techspectations held at LeMeridien in Kochi on Friday.

Natarajan said entrepreneurs should be wary of the misinformation and misguided data that gets into the system when AI is deployed at enterprise scale.

“Every inference that is coming out of AI, we should know where it is coming from,” Natarajan said.

Praveen Jayakumar, Senior Manager AI/ML, Amazon Web Services, gave the example of the latest blockbuster chatbot, ChatGPT, to drive home the point of misguided ingestion of data.

“ChatGPT is trained to give an answer, which is not always the correct answer. It gives answers in quite an authoritative fashion, but these may not be right. That's the danger,” Praveen said.

Already, ChatGPT's answers have been revealed as having certain inherent biases especially concerning race and colour that are not in keeping with how the civilisation would want itself to evolve.

“My biggest fear is that the underlying data with which you train the AI model, if it has biases like race or colour, will be carried over to systems powered by AI,” Praveen said. In short, AI can colour enterprises in ways that would not be appreciated.

Is AI really smart?

A way out of the moral damage could be an oversight mechanism. “There are enough technologies to ensure explainability, fairness and transparency in AI systems. Just make sure there are enough of these in the AI system you adopt,” Subram said.

Also, he said there should be a governance mechanism within the organisation to inspect that things were being done in alignment with the principles adopted.

Soumitra Dhankar, the head of Solution Consulting, Adobe India, said that there was a difference between “sounding smart” and “being smart”. He said the adoption of AI could make enterprises sound smart. “But companies will have to ensure that there is enough accountability built into the system,” Dhankar said.

He said a company's intent and emotion mattered. “A company has to pledge that it would respect the privacy of its clients and customers,” Dhankar said.

AI is here to stay

Nonetheless, the advance of AI is not something that could be reined in. Amazon's Praveen said that there were five ways in which AI and machine learning were making its presence felt. One, model sophistication. The models are getting larger and are developed using massive amounts of data.

Two, the employability of an increasingly wide array of data. “Machine learning right now is good at identifying things like a photo or a video. Take a football game, for instance. It can identify what is happening on the field and who is playing. But when and where are the big questions that are still left unanswered. This is now being developed through the evolution of geospatial data,” Praveen said.

Three, the standardisation of AI infrastructure tools. Four, responsible AI. “Lot of government regulation is in the offing. Already, the European Union has issued guidelines on bias,” he said.

Five, democratisation. Lot of work is happening to take AI out of the hands of just top global universities and some big companies.

Why do we need AI

Dhankr of Adobe cited three reasons why we need AI. One, “to discover wisdom”. “We are struggling to convert data into information, filter information into knowledge, and create wisdom out of knowledge. Essentially, how do I discover wisdom out of knowledge,” Dhankar said.

Two, “accelerate what is now slow”. Dhankar gave the example of movie production. A filmmaker wants to shoot a scene in the morning. But his lead actor is available only in the afternoon,and others in some other time. “So the filmmaker has to shoot his morning sequence in different time zones. Can AI help him effectively mask the pixel differences of various time zones? In other words, can AI quicken this seamlessness in production,” he said.

Three, decide whether AI matters. Dhankar said humans have upstaged goldfish as the species with the shortest attention span. If goldfish's was nine seconds, humans have recently logged eight seconds. “Therefore, the idea is to engage. How do I keep my customers engaged,” he said.