eweekchat

A7. Inclusive AI needs to be a top priority for all those in the industry, and it starts with the data. We are seeing responsible AI gain a lot of traction, with different tools being used to detect bias in the data or with the model.

A7 New quality standards for AI systems are being shaped as we speak. It is important for the AI engineers to see standards and best-practices being established. For instance the ISO-29119 defines what characteristics of an AI system should be tested.

@unsolvedmistry I've heard that from several executives. "Inclusive AI" might the watchword of 2021 AI.

I agree with @unsolvedmistry. Inclusive and Trustworthy AI should be the norm. We need AI to improve our society not repeating the mistakes of the past.

A7. @JamesMaguire #AI & ppl should be more tightly connected. In how it's developed and how it's used. It learns from us and in the end ppl should be learning from the solutions they implement.

A6. We were surprised to see the “silver bullet” narrative less and less, and how organizations have shifted to using AI to support IT operations and improve internal efficiencies. People are seeing success in limited-scope use cases, as opposed to more general approaches.

A6. I'm not sure if it's a surprise but we see more enterprises building AI centers of excellence, consolidating dev silos, centralizing expertise, process, infra

A6:
1) Responsible or Trustworthy AI is becoming the norm and a key business differentiator,
2) New professions such as AI Explainer, AI Controller are being established,
3) 7 out of 10 AI projects are being successfully deployed. Now the percentage is less than 30%.

@TonyPaikeday So an "AI center of excellence" is a dev center for AI? Or what?

A6. Software development becomes an integrated discipline with AI/ML development.

A6 @JamesMaguire yes - where an enterprise was previously seeing a lot of "shadow AI", IT taking the mantle and building centralized platform for model prototyping/training at scale. Also better workflow tooling ie: MLOps

A6. @JamesMaguire automation & the idea of self-healing tasks is trending. The idea is for a network to fix itself as IT teams may not be around in our #hybrid workplaces.

A6: The "mllestone" I see, if it can be called one, is that some form of AI or ML is being added as a feature or a core component into virtually every new business app that becomes available. Or so it seems anyway--I don't have metrics on that.

A5: There are few leaders. @nvidia @awscloud and @google are very strong in #AI #eweekchat

I see the cloud hyperscalers as key providers, in that their customer base is already hooked into them.

A5. From our clients we see that they tend to use their existing cloud providers (Microsoft, IBM and AWS mostly) as their top platform providers or key vendors. Then, they choose from a plethora of open source tools out there to complement their infrastructure.

(edited)

A5. There is a correlation between budget and market leadership perception. You can’t be an AI leader without consistent budget investment.

@zkerravala @nvidia is particularly hot, in my view.

A5. @JamesMaguire In our space we're seeing the cloud providers always mentioned as ppl assume that holding the data means its a platform.

A4: A first good step is to define and scope very well the business problem at hand that you expect that AI will solve.

A4. @JamesMaguire Everyone in the org - top down - should determine their goals, social the changes and advantages, and then see what solution or vendor aligns with the task at hand.

@trentf Sounds like it could be a 6-12 month process?

@ykanellopoulos The right problem to solve can be the biggest hurdle for an organization. It's critical that they define and scope upfront.

A4. When looking to expand AI deployment, look for a partner who has done this before. We have found that those working with external data providers are 1.5x more likely to be ahead of others in AI adoption.

A4. @JamesMaguire In the network space its a shorter timeframe as #AI is bundled into management for them. But I've seen teams take a while deciding on how to handle change. 8-)

A4. I agree with @unsolvedmistry. First set the problem right and then get the right team together. Be open to external expertise.

A4. from am infra scaling perspective we see many struggle with where to place compute resources. Consider role of both cloud and on-prem. Many orgs are repatriating AI workload from cloud in order to bring training closer to where their data lives.

A3. As in every aspect of IT, I’d say yes, there is lack of AI talent. We need more skilled professionals, this time with a diverse skillset from engineering to classical/humanitarian studies.

There are jobs open everywhere in IT, especially in administration for AI and cybersecurity. Companies are looking everywhere for candidates to fill these jobs.

A3 The lack of skilled professionals is the biggest bottleneck for AI. With emerging no-code/low-code, AutoML solutions out there, as well as courses like AI for PMs on @udacity that we created, there’s hope we can bridge that gap.

(edited)

@unsolvedmistry Very true, Sid. Hopefully the resurgence of low/no code development will bring more people to these good-paying positions. As long as the processes and steps are clearly laid out, the drop-downs and wizards work, people should be able to do these jobs.

@editingwhiz Be careful with the low/no-code development as in practice it is being proved to be insufficient to cover the needs of large scale systems. It is mostly valuable for well-scoped small to medium systems/applications.

@ykanellopoulos You right, of course, but no/low-code development can also serve as a good onramp to development in general.

@ykanellopoulos I'm guessing there are some low code/now code advocates who would disagree.

I am only a data-point in this sea of opinions :-). In general it is good to set expectations right when it comes to no/low code solutions. I have seen some good examples and some big failures.

@editingwhiz Totally agree.

A3. @JamesMaguire There is so much #AI happening that talent is spread thin. There's also a challenge on the implementation side. Something new takes training & that's an issue as well.

A2. So much of the AI intellectual property companies develop still often never realized in production applications. I think many organizations lack IT processes and workflow management tooling to move models from prototypes into production. "operationalizing AI"

A2. Organizations need to pick the right first problem to solve, often called the Goldilocks problem of AI. It can be a challenge to determine what problem and size of problem to solve first.

@TonyPaikeday So it's almost an "expensive science experiment"

A2.
Lack of trust
Lack of skill
Lack of standards/soft-infrastructures (call me proper data/datasets)

(edited)

especially based on how much you're paying your data science team :)

A2. also many undeployed AI models that languish in prototype because many companies lack platform and devops rigor for taking a viable prototype into scaled training into production inference/deployment.

A2. @JamesMaguire Fear of the unknown isn't helping with adoption. I just heard an analyst mention that ppl assume AI will take their jobs. A myth we need to help dispel.

@trentf But some jobs will be lost, don't you think?

Jobs and the expectation of ppl need to morph. I understand that change is challenging, but the mindset should be about finding opportunity.

I hear execs complain that it's not ready for prime time....

A1. On average I’d say it’s moving away from its infancy but it has some way to become mainstream. However, for certain domains such as image recognition/computer vision it’s moving towards mainstream.

A1. I think where AI was squarely anchored in science and academia a few years ago, we now have pragmatic use cases in every vertical supported by proven models for everything from recommender systems to natural lang processing to anomaly detection

@TonyPaikeday
Is one vertical strongest in your view?

(edited)

A1. AI is now critical to business success. Based on @appenglobal 2021 State of AI report coming out June 15, we are seeing significant increasing budgets, and responsibility for AI initiatives move to ops levels. This signals AI is maturing within organizations.

A1. I've seen a few good examples - hospitals use it to speed clinical image processing from 5min down to 2sec. Domino's pizza using AI to improve order readiness prediction accuracy from 75 to 95%. even seeing it used to improve accuracy of detecting bovine tuberculosis

@TonyPaikeday If it makes pizza faster, I'd say it's succeeding :)

A1. I also see image recognition startups using AI for substituting nurses for night-shifts.

(edited)

A1. Also - many companies can only mine say 2% of the call center recordings for helping train agents. That 2% is transcribed with maybe 20%-40% accuracy, but AI is letting companies mine 100% of their recordings with almost 100% accuracy

A1. Trent from @arubanetworks @JamesMaguire The AI market is in different phases for different use cases. While $ is being budgeted IT teams are just starting to embrace it in the network space.

A1. Another area we see constant adoption of AI is cyber-security, especially using unsupervised techniques.