eWeekChat

A5- Waiting. The advantage AI gives is incredible and at some point will become insurmountable by those who don’t deploy.

A5. Companies must avoid “AI for AI’s sake” and first establish the right use case application to guide their AI journey.

A5. Training a model is not the end of the ML lifecycle but the beginning. The world changes, sometimes quickly, and an ML model can go stale fast if left untended.

A5: AI with big data is a complex engineering and systems problem. Paying attention only to AI research, while ignoring the engineering, data and systems challenges will only lead to great POCs and not great products.- G.T.

(edited)

Bringing everyone along. the business win is always good, but at some point there will be a challenge. Feeling collective success tends to get over those humps with far more ease than going it alone.

@dgspjs +1 -- love this point. Models cannot / should never be considered static. If we're modeling human behavior, we have to remember that humans learn. So to must the model.

@sunil_senan - reminds me of "what is your linux/XML/etc. strategy?" No! no! no! These are tools, not solutions.

A5: Developing so-called AI tech without product development is an experimental exercise in data management and software development.

A4 - Start small. Then scale with familiarity and comfort....Sorry I am kind of repeating myself

A4P2 Also really needs to get that needs analysis solid before starting..

@brianpodolak Makes perfect sense. I think "start small" is industry accepted wisdom at this point.

@JamesMaguire You would be surprised!

(edited)

A4. Use model and prediction explanability to understand how your model works. Dramatically reduces the risk of model failures and can be a source of major insight into the business.

Program management 101... narrow focus, quick wins. Also look for use cases where there is a lot of data.

A 4. Creating an #AI road map to identify possible implications on the people, process, and technology – the challenges they will face, and possible resolutions they can turn to.

A4: Getting their Data Strategy, Data Quality, Data OPS right. AI is only as good as the underlying data it is built on. Investment in Data maximizes AI ROI. - G.T. #eWeekChat

(cont'd) e.g., applying AI to transactional data is usually interesting because every company usually has a lot of it

@steveshah Love this thought, and opens a huge can of worms...but very exciting for sure!

A4: Well-staffed and well-funded data science, data management, and software development are critical to developing true AI innovation.

A3 -Getting their head around it, first. Then when employees see it, it becomes a threat to their livelihood. A lot of "flying unicorn" expectations.

A 3. Scarcity of talent, risk-averse cultures, and an inability to imagine what the solution will look like on completion.

@sunil_senan Scarcity of talent is a big one. Price of talent is the problem.

A3. Getting the right data for the desired business case is always #1.

Transitioning models (and their data transformation pipelines) from training to production can be an Ops nightmare.

Diagnosing issues with black box models when something goes wrong. Kno

Even knowing that something in a production model has gone wrong.

Agree with Brian. Setting expectations is key. A lot of end consumers of the technology still have an impression of AI based on the movies. The other side is expectations - AI isn't magic, but it is patient.

A3P2 - I think you should also implement in stages..baby steps..don't try to replace a complete process...start small.

(edited)

@steveshah "AI based on the movies" is very real. It's magic! Though it's not....

@sunil_senan These are business drivers for sure. The desire for AI from leadership is key. Getting the broader organization on board has to happen too

Totally agree. I still run into a lot of "AI as magic wand."

:-) Per Short Circuit... "It doesn't get happy, it doesn't get sad... it just runs programs."

A3: They believe ML is true AI when it is not. They need to learn how to apply ML at scale organization-wide before materially tackling deep learning (DL) applications and human-like AI.

(edited)

Also nurturing that talent within business and IT is equally important. Digital transformation and talent transformation go hand in hand.

@steveshah Completely agree. This is key to drive broader alignment and culture across the enterprise, but it also helps to drive AI adoption and prioritization to solve meaningful problems.

A2 - Employment, inflation and costs are the biggest driver for interest. And of course, "Noone wants to work"

(edited)

A2. Conversational AI (AI assistants, voice AI, avatar-based assistants), hyper personalized experiences, increased adoption of low-code no-code (#LCNC) platforms, domain-specific AI models.

@sunil_senan I find that conversational AI is very hot now.

I'd agree, but humanized conversational is the future...

A2. MLOps automation & ML fairness.

Managing models and data pipelines once things reach production is no small feat and can suck lots of time from some of the most in-demand labor around. Automate that. 1/

Our Now and Next report shows 90+% of business leaders report automation helps address staffing shortages, supply chain issues, and the ability to meet customer requirements https://bit.ly/now-and-next-report

A2p2. AI can be harmfully biased if care is not taken in their creation and regulators are starting to pay attention to that. Beyond potential fines, there is huge reputational risk in deploying harmful models. Audit your models or someone else will! 2/

@brianpodolak "humanized conversation"? So it's an AI category?

@DGSPJS Very true!!

@JamesMaguire Should be....I think it will be....Conversational is too stoic/robotic in most implementations imho

(edited)

@brianpodolak - Conversational AI is definitely interesting, albeit temperamental. More broadly, I would say packaged solutions are key. They drive ease of consumption - I don't have to be plugged into the how to use.

@steveshah @brianpodolak I agree that packaged is the way to go - the only way for more firms

@steveshah If you can find a package that meets your needs. Most of our clients have been implementing hybrid or custom solutions though.

A2: The use of machine learning (ML) to convert data into business outcomes is the chief driver of the AI market, with corporate AI focus, spending, and applications based on the ML submarket.

Indeed James. It opens up a number of use cases that enable businesses to operate with digital clock speed.

A1 - Both. Ai is layering in. It is a software. That can be buggy at times especially in the beginning.

A1. Very mixed bag. Healthy data practices makes or breaks any AI deployment. Way beyond model architecture or anything like that, successful usage demands a robust data infrastructure. Those who get there are seeing incredible value.

A1. Businesses are facing daunting challenges to scale #AI. Infosys’ new Data and AI Radar report reveals that only 26% of practitioners are highly satisfied with their data and AI tools.

Everyone is getting their arms around AI with interesting use cases. The edges of appropriate are key as are the tie-ins with rest of the ecosystem (esp. automation tools).

@sunil_senan Why so low, do you think?

We're still on the on-ramp. Good vertical tools are key. e.g., document processing

Sounds right to me. Largely immature tooling, lots of people rolling their own from open source and finding rough edges.

@dgspjs And spending way more than they hoped!

A1: They’re getting started in identifying AI’s potential and use cases and frustrated with how far they are from true AI. How they get there and scope are unclear.

Businesses are struggling on 3 counts: 1. Not having a clear use case & outcome, 2. Data quality continues to be a big issue, 3. Not involving business executives in the process hinders adoption and trust in AI