Data vs Models: The AI battle of the future

November 11, 2022

AI needs good data. But data and the insights created are often unwieldy and hard to find, let alone define. The first article in this newsletter looks at the need for a holistic data strategy, with a pyramidal approach wherein a few “superstar” data scientists are supported by a wide base of citizen “business” data analysts, enabling insights to be created at speed that traverse all business functions.

We also look at how open data is transforming the EV industry, and give insights into how a connected vehicle platform with telemetry can be built from the ground up. Our third article is our most popular AI article thus far, and gives insights into how tiny AI models are upending how computer processing takes place on the cloud continuum. We also offer insights into a new breed of “foundation” models, which are able to take advantage of the exponential increase in compute power to write literature, paint, and write poetry (with more aplomb than many humans). AI in healthcare is also a prescient subject, and we explain how augmented intelligence will enable doctors and physicians to provide holistic, personalized healthcare solutions that are unique to each individual, no matter their tale or creed.

The need for a better data strategy

Good use of data can make firms more competitive, help them save money, improve business processes, increase customer-centricity and develop ground-breaking marketing strategies, as Google, Netflix and Walmart have shown.

But there is a difficulty at most enterprises. A data driven culture is lacking. Skills in data analytics are in short supply. And there is a lack of experience in applying data analytics to business issues.

In this short paper, we propose four steps to a holistic enterprise-wide data strategy. One of those steps is to build a new operating model wherein data scientists are at the top of a pyramid, with “citizen scientists” and business executives with some knowledge of data models and implementation challenges forming the base. These data-driven executives lead by example and enable the firm to rely less on rock-star data scientists, democratizing insights throughout the organization.?

Birthing a data-centric EV ecosystem

How do you give EV owners more control over their data, while also spurring an innovative ecosystem of OEM’s, app developers, and other data end users?

The secret is a connected EV platform with open telemetry devices in the car itself.?

Characteristics of this platform also include data sources from IoT and other edge devices, trusted execution environments, economic incentives for sharing data, and real-time traceability, simulation, AI-based decision, and virtualization of experience and operations.

The goal here is to establish trust and transparency across multiple untrusted/semi-trusted parties, individuals or smart devices and overcome some of the technical challenges through clever use of partnerships.?

The future of AI is Tiny

How can we create a business paradigm where AI systems are private, highly responsive, contextually intelligent, eco-friendly, and talk to each other in real time?

Enter Tiny AI – a future where big enterprise will infuse edge intelligence in customer-centric digital engagements that retain value at scale.

This is one of applied AI’s most advanced trends yet. Here, complex algorithms run on edge devices themselves, all without the need for internet connectivity.

But there is a problem. Not all AI will work on the edge, and for many, applications and models still require the computational resources and processing technology provided by massive cloud data centers.

With the problem, comes a host of solutions, including clever AI techniques such as data and model engineering, MLOps for Edge, along with device-specific tuning, deployment, and validation.?

Are foundation models the future of general AI?

AI models not only need to be explainable and responsible, but flexible. Ideally, they should be reassigned from one problem to another with relative ease by means of fine tuning. Enter the “foundation model”, changing not only what AI can do but how AI works as a business.

Oren Etzioni, founder of the Allen Institute for AI, a research outfit, estimates that more than 80% of?AI?research is now focused on foundation models—which is the same as the share of his time that Kevin Scott, Microsoft’s chief technology officer, says he devotes to them.

Such models are trained using a technique called self-supervised learning, rather than with pre-labelled data sets. As they burrow through piles of text they hide specific words from themselves and then guess, on the basis of the surrounding text, what the hidden word should be. After a few billion guess-compare-improve-guess cycles this Mad-Libs approach gives new statistical power to an adage coined by J.R. Firth, a 20th-century linguist: “You shall know a word by the company it keeps.”

Good enterprise AI is more about data than models

Employing humans to label data for AI models is expensive, and there is no guarantee that output will be comprehensive, unbiased, and free from noise. It also takes a long time, a problem for firms that aim to be Agile, and that release new products and updates week to week.

In fact, at Infosys, we believe that up to 60% of ML project costs go toward manual labelling and validation.

Is there a better way to improve AI throughput and reduce costs?

In fact, there are a number of ways, from “intelligent learners” that proactively pick parts of the data for human labelling, to completely synthetic data creation, the kind that Amazon have deployed in their Amazon Go stores, and Nvidia, the chip maker, are proposing for self-driving cars. Some even say that these completely made-up datasets enable AI models to work better than those trained on the real-world alone.?

The use of AI in Preventive Healthcare

How can AI be applied in healthcare?

For many healthcare providers, AI can be used as a tool to transform the approach of linear patient journeys to a patient-centered 360-degree methodology that promises more preventive, personalized and outcome-based medicine.?

Prevention is the holy grail. Personalization too. Ultimately, AI will make disease treatment radically more cost-effective by personalizing care to each person’s unique condition and experience, and by treating the causes and symptoms of disease.

To get there, good use of cloud is necessary. AI must be deployed across the full healthcare ecosystem and integrated with partners in every step of the cycle of care. Other necessary conduits include responsible AI and explainable AI, with human hands governing how AI is used.?

To continue to receive Infosys Knowledge Institute newsletters, please do follow us on LinkedIn and subscribe to the newsletter. Please do share it with your teams and your professional network.