I sat down with Teresa Tung to larn much astir the changing quality of information and its worth to an AI strategy.

AI occurrence depends connected aggregate factors, but the cardinal to innovation is the prime and accessibility of an organization’s proprietary data. 

I sat down with Teresa Tung to sermon the opportunities of proprietary information and wherefore it is truthful captious to worth instauration with AI. Tung is simply a researcher whose enactment spans breakthrough unreality technologies, including the convergence of AI, information and computing capacity. She’s a prolific inventor, holding implicit 225 patents and applications. And arsenic Accenture’s Global Lead of Data Capability, Tung leads the imaginativeness and strategy that ensures the institution is prepared for ever-changing information advancements.  

We discussed a big of topics, including Teresa’s six insights.

Finally, we concluded with Teresa’s Advice for concern leaders utilizing oregon funny successful AI 

Susan Etlinger (SE): In your caller article, “The caller information essentials,” you laid retired the conception that proprietary information is an organization’s competitory advantage. Would you elaborate?  

Teresa Tung (TT): Until now, information has been treated arsenic a project. When caller insights are needed, it tin instrumentality months to root the data, entree it, analyse it, and people insights. If those insights spur caller questions, that process indispensable beryllium repeated. And if the information squad has bandwidth limitations oregon fund constraints, adjacent much clip is needed. 

“Instead of treating it arsenic a project—an afterthought—proprietary information should beryllium treated arsenic a halfway competitory advantage.”

Generative AI models are pre-trained connected an existing corpus of internet-scale data, which makes it casual to statesman connected time one. But they don’t cognize your business, people, products oregon processes and, without that proprietary data, models volition present the aforesaid results to you arsenic they bash your competitors.   

Companies put each time successful products based solely connected their opportunity. We cognize the accidental of information and AI—improved determination making, reduced risk, caller paths to monetization—so shouldn’t we deliberation astir investing successful information similarly? 

SE: Since truthful overmuch of a company’s proprietary cognition sits wrong unstructured data, tin you speech astir its importance? 

TT: Yes, astir businesses tally connected structured data—data successful tabular form. But astir information is unstructured. From dependable messages to images to video, unstructured information is precocious fidelity. It captures nuance. Here’s an example: if a lawsuit calls lawsuit enactment and leaves a merchandise review, that information could beryllium extracted by its components and transferred to a table. But without nuanced inputs similar the customer’s code of dependable oregon adjacent curse words, determination isn’t a implicit and close representation of that transaction.  

Unstructured information has historically been challenging to enactment with, but generative AI excels astatine it. It really needs unstructured data’s affluent discourse to beryllium trained. It’s truthful important successful the property of generative AI. 

SE: We perceive a batch astir synthetic information these days. How bash you deliberation astir it? 

TT: Synthetic information is indispensable to capable successful information gaps. It enables companies to research aggregate scenarios without the extended costs oregon risks associated with existent information collection.  

Advertising agencies tin tally assorted run images to forecast assemblage reactions, for example. For automotive manufacturers grooming self-driving cars, pushing cars into unsafe situations isn’t an option. Synthetic information teaches AI—and truthful the car—what to bash successful borderline situations, including dense rainfall oregon a astonishment pedestrian crossing.  

Then there’s the thought of cognition distillation. If you’re utilizing the method to make information with a larger connection model—let’s say, a 13-billion-parameter model—that information tin beryllium utilized to good tune a smaller model, making the smaller exemplary much efficient, outgo effective, oregon deployable to a smaller device. 

AI is truthful hungry. It needs typical information sets of bully scenarios, borderline conditions, and everything successful betwixt to beryllium relevant. That’s the imaginable of synthetic data.   

SE: Unstructured information is mostly information that quality beings generate, truthful it’s often case-specific. Can you stock much astir wherefore discourse is truthful important?   

TT: Context is key. We tin seizure it successful a semantic furniture oregon a domain cognition graph. It’s the meaning down the data. 

Think astir each domain adept successful a workplace. If a institution runs a 360-degree lawsuit information study that spans domains oregon adjacent systems, 1 domain adept volition analyse it for prospective customers, different for lawsuit work and support, and different for lawsuit billing. Each of these experts wants to spot each the information but for their ain purpose. Knowing trends wrong lawsuit enactment whitethorn power a selling run approach, for example. 

Words often person antithetic meanings, arsenic well. If I say, “that’s blistery for summer,” discourse volition find whether I was implying somesthesia oregon trend.  

Generative AI helps aboveground the close accusation astatine the close clip to the close domain expert. 

SE: Given the gait and powerfulness of intelligent technologies, information and AI governance and security are apical of mind. What trends are you noticing oregon forecasting? 

TT: New opportunities travel with caller risks. Generative AI is truthful casual to use, it makes everybody a information worker. That’s the accidental and the risk. 

Because it’s easy, generative AI embedded successful apps tin pb to unintended information leakage. For this reason, it’s captious to deliberation done each the implications of generative AI apps to trim the hazard that they inadvertently uncover confidential information. 

We request to rethink information governance and security. Everyone successful an enactment needs to beryllium alert of the risks and of what they’re doing. We besides request to deliberation astir caller tooling similar watermarking and confidential compute, wherever generative AI algorithms tin beryllium tally wrong a unafraid enclave.  

SE: You’ve said generative AI tin jumpstart information readiness. Can you elaborate connected that? 

TT: Sure. Generative AI needs your data, but it tin besides help your data.  

By applying it to your existing information and processes, generative AI tin physique a much dynamic information proviso chain, from seizure and curation to consumption. It tin classify and tag metadata, and it tin make plan documents and deployment scripts.  

It tin besides enactment the reverse engineering of an existing strategy anterior to migration and modernization. It’s communal to deliberation information can’t beryllium utilized due to the fact that it’s successful an aged strategy that isn’t yet unreality enabled. But generative AI tin jumpstart the process; it tin assistance you recognize data, representation relationships crossed information and concepts, and adjacent constitute the programme including the investigating and documentation. 

Generative AI changes what we bash with data. It tin simplify and velocity up the process by replacing one-off dashboards with interactivity, similar a chat interface. We should walk little clip wrangling information into structured formats by doing much with unstructured data.  

SE: Finally, what proposal would you springiness to concern and exertion leaders who privation to physique competitory vantage with data? 

TT: Start present oregon get near behind.  

We’ve woken up to the imaginable AI tin bring, but its imaginable tin lone beryllium reached with your organization’s proprietary data. Without that input, your effect volition beryllium the aforesaid arsenic everyone else’s or, worse, inaccurate. 

I promote organizations to absorption connected getting their integer halfway AI-ready. A modern integer core is the exertion capableness to thrust information successful AI-led reinvention. It’s your organization’s premix of unreality infrastructure, information and AI capabilities, and applications and platforms, with information designed into each level. Your information foundation—as portion of your integer core—is indispensable for housing, cleansing and securing your data, ensuring it’s precocious quality, governed and acceptable for AI.  

Without a beardown integer core, you don’t person the proverbial eyes to see, encephalon to think, oregon hands to act.  

Your information is your competitory differentiator successful the epoch of generative AI. 

Teresa Tung, Ph.D. is Global Data Capability Lead astatine Accenture. A prolific inventor with implicit 225 patents, Tung specializes successful bridging endeavor needs with breakthrough technologies.   

Learn much astir however to get your information AI-ready: 

• Get to cognize the Microsoft Intelligent Data Platform. 

• Learn wherefore a beardown information and AI instauration is essential to each organization’s success, contiguous and tomorrow.  

• Learn however to make an intelligent information strategy that endures successful the epoch of AI with the downloadable e-book. 
• Watch this on-demand webinar to perceive Susan and Teresa spell deeper connected however to extract the astir worth from information to differentiate from competition. Learn astir caller ways of defining information that volition assistance thrust your AI strategy, the value of preparing your “digital core” successful beforehand of AI, and however to rethink information governance and information successful the AI era.

Visit Azure Innovation Insights for much enforcement position and guidance connected however to alteration your concern with cloud.