2 days ago
Shepherding Data To Its Eventual Destination
What does it mean to engineer data transfers in the age of AI? More than a few top professionals would say that this sort of thing is even more important now than it had been in the past. After all, artificial intelligence works on the principle of taking in information and processing it – in ways that are eerily similar to what happens in our human brains.
So the information itself is valuable, and the processes are valuable, too.
Also, modern advances in large language models have given us different ways to look at data transfer. I'll get into some of that with thoughts on three lightning talks given at Imagination in Action in April, in the Lightning Talks section of the wide-ranging program.
Unlocking Siloed Data
The first such presentation was from William Lindskog Munzing who was talking about an application called Flower, suggesting that traditionally, data has been trapped in siloes.
The goal, he said, is to move AI to where the data is. That's much easier in times when work on quantizing and foundation models and lower-bit systems has ushered in edge AI – the ability to locate the AI where the data already is, rather than porting it to centralized data centers.
With that in mind, the Flower community numbering some 5800 developers, with 2000 active projects, is working on what Munzing calls the 'Collective-1' user-owned platform.
'What we believe in is that data should stay where it originates,' he said. 'It's … never transmitted. It stays in your device, in your car, whatever it is, or in the hospitals.'
The ISO-certified project, he added, is also versatile.
'We have done a lot of things in the deployment runtime,' Munzing explained. 'So it's very easy for you now to run your experiment on CPUs, GPUs, and then scale it to actual real world deployment if you want to add secure mechanisms, authentication and much more.'
AI for Leads
The next talk came from Marco Cello, who worked on a project called Meshify.
He explained that data suggests small to medium sized enterprises (SMEs) are on average about 50% as efficient as corporations, and that collectively, SMEs lose up to $500 billion in revenue from poor lead management alone.
As a solution, Meshify, he explained, will scour a professional's inbox, follow up, and provide them with automated CRM insights.
I thought it was also interesting that this project uses the NANDA decentralized network idea pioneered at MIT by some colleagues of mine. Actually, so does Flower, which indicates the initiative to create a functional web protocol for AI is picking up steam.
Regulation and Control
The third presentation came from Peeyush Aggarwal at Deloitte, where he talked about dimensions of change in the AI era.
This was different, because Aggarwal wasn't promoting a particular startup or product. Instead, he was talking about cycles – a cycle from assistive to augmentation to automation, and priorities for AI development, including:
'Human strengths need to be amplified,' Aggarwal said. 'How do you take a banker and train them to do cognitive, as well as decision making, when people are designed to think in a straight line, right? And it's about that ability to negotiate, to be able to take decisions ethically in an environment, (this) is the most important aspect.'
Referencing a need to control the culture of change in the banking industry, Aggarwal went over many aspects of analyzing AI activities from a boardroom perspective, citing a gradual change and evolution from classic to digital banking, and then to intelligent banking.
At the end of his presentation, he went into a sort of an interesting philosophical path, talking about the management of human and AI agents.
'The most important part is, when you bring human and AI agents together, is, do you control the AI, or do you control the decision of AI and humans coming together?' he asked. 'That's the most important aspect. What's the use case? When a regulator looks at a use case, he asks the question, 'Can I repeat that use case? Can I repeat the question that is being asked?' And if you can't, you can't really approve their use case.'
In other words, managers who are managing human and AI agents have a different role than those who are managing only humans themselves.
You're managing the intersection of humans with technology. How does this work?
I thought these were some interesting eye-openers in a time when we're trying to adjust to a rapidly changing target in terms of technology use. Stay tuned.
