eWeekChat

A9: Fail fast with AI.
This is a different mindset that’s different from traditional software. You need to deploy your AI in production so that you can learn quickly and make the appropriate adjustments. All the trainings in the lab will do you no good. That’s not how AI works.

A9: Overall, #AI - and particularly #AIOps - helps #IT Operations lead their company’s evolution into an autonomous digital enterprise - embracing intelligent, tech-enabled systems across every facet of the business.

@victorthu love it! FAIL FAST, succeed faster. Get back up and try again. The old adage of fall off your bike and get on to ride again is true for AI.

@RyRaiker Of course "fail fast" can be pricey with AI....

A9: as we started to touch on in the last question, companies should be more aware of ethical use of AI, and how their data may be used by some companies that are not disclosed, such as for training their AI models.

@victorthu A6: I think my answer is along the same lines as Victor's. I wouldn't call it 'failing fast' though as I think the harms and risks are amplified with AI (than even traditional software and the harms weren't insignificant there).

it doesn’t have to be - composable AI or composable apps means things that were built once can be reused and redeployed. No time wasting when you’re improving. I think the problem some have is they get so far along, but from the beginning they never truly understood

A9: The ethical and socioeconomic impacts of AI — in parallel with tech designed to mimic/surpass human capability, far beyond a utility. It is in corporate fiscal interests for this to be less of an unknown.

A6: I think finding quick paths to production, working transparently and getting many eyes on AI-enabled applications is important. It's boring but I think robust ML practices make all these things easier

Yes @JamesMaguire, 'fail fast' can be expensive. But it's more important to get going and "monitor and govern" them closely. Real world data is essential for models to learn and improve.

A9: There still seems to be a dearth in data science expertise. So we need more expertise. Lots of students getting trained up now and some autoML tooling will help to get to a reasonable model faster but still early days.

A9: It's not so much about "failing fast", but learning fast and incorporating learning into both your models and processes.

@RikChomko Very true on the dearth of expertise. Which drives up the cost, of course.

on the topic of failing… when I started driving, I didn’t take the car to 100mph, I started slow on the side streets. Moreover I was quite ok when I hit the curb because I wasn’t at high speed. The same could be said for AI projects - follow the speed limit

A6: If, for example, some significant harm is discovered then getting the offending model offline in a way that doesn't break the overall solution, figuring out how to provide redress if necessary are all solution design-related issues

AI has so much potential, saving time, money, and changing the world. Isn’t it time we ask why not rather than sit stagnant?

A6: On the rapid identification of harms I would love to see more intentional partnering between orgs and civil society auditors. More algorithmic bug bounties a la Twitter's META team for example.

A8: The economy today is a good forcing function for enterprises to stop treating AI like a toy.

In 3 to 5 years, enterprises who pivoted from building their own tools will advance much rapidly as they focus on delivering real results rather than just doing lab research.

A8: The market and competition will solve ML. Innovators will begin to properly position themselves in the DL and true AI segments.

A8: There is a lot of potential for #AI – especially its predictive capabilities. Additionally, we expect to see AI ethics standards proliferate among the tools & consortia as governance around AI starts to codify.

A8: It will certainly grow but it will need to adapt to be more explainable and proactively alert organizations for bias in their models/data.

A8: My prediction/hope is modest. I think we'll see more standardised practice around MLOps and DataOps. And I think a corollary of that is an easing of the deployment gaps so many orgs are currently grappling with.

@BCorrIoT "governance around AI" -- that's a big topic.....

A8: I guess we will see iRobot come to life… no. No. No! I suspect you will see remote work grow, while production increases. As inflation continues, more companies will be looking for cost savings which AI can deliver. Those who remain stagnant will be beat!

@RyRaiker Good point on inflation driving AI adoption. Clearly yes.

This is a really good point. I think that governance is both internal and external (i.e. from regulators and civil society-based auditors)

Yes @JamesMaguire we are seeing this in conversation with both private and public sector customers.

@BCorrIoT This is a really good point. I think that governance is both internal and external (i.e. from regulators and civil society-based auditors)

A8 (continued) I think these AI tools will be driven by the rise and commoditization of process mining which will mean process and data understanding for every business operation and tech stack.

@BCorrIoT Still on the governance front. I really hope regulation keeps up. I think it's a useful forcing function for building better, more robust AI.

Exactly @Adewunmi , in fact this is critical for AI to gain wider adoption.

@victorthu but why hasn’t it been seen yet?

@Adewunmi Regulation varies from region to region. For example the EU is getting ahead of this problem right now.

@Adewunmi I'm pessimistic about AI governance. Too much money, too many ways to work around regulators. And what's their authority?

@RikChomko Yeah, I agree. I think the path to this is a holistic one though. I don't think we'll see tools emerge that do this well i.e. no silver bullet. I think better ML practice is at the root of this and the cost and pain of maintenance may help drive this too.

@Adewunmi the EU has done a really amazing job of looking at these angles https://www.abbyy.com/blog/legal-regulation-of-art...

A7: Specialists that solve a level of the tech — ML, DL, AI — with clear applications and business cases.

A7: The big cloud guys have built a series of complex tools that require expertise to assemble. Like supermarkets offering just anything as long as you have chefs who knows how to cook.

Instead, the more exciting players are the startups that are providing gourmet meals!

Q7: it might not be who you suspect. But the best way to evaluate leaders is through analyst reports. AI is an umbrella tech that covers a variety of enterprise tech. and it plays a huge role in intelligent process automation.

@victorthu You mean low code no code?

A7: Well, for obvious reasons I would say Cloudera :). One reason for that is that data is so critical to delivering high-quality AI. Making data ingestion as easy as possible regardless of where that data sits - on prem or in the cloud, is the first step

@RyRaiker on the note of analysts : In RPA for example, Gartner has recognized UiPath, BluePrism and Microsoft as leaders.

A7: @BMCSoftware is a proud leader in this market & deeply committed to #AIOps. Our AIOps solutions apply #ML & predictive capabilities across IT Ops & #DevOps environments for real-time, enterprise-wide observability, insights, & automated remediation. (1/2)

Not exactly @JamesMaguire. There are tools offered by startups today that allow AI models to go into production quickly without having to assemble your own infrastructure.

@RyRaiker For intelligent document processing, Everest Group has named ABBYY and Automation Anywhere as leaders, and in process mining NelsonHall recognizes ABBYY.

A7: I also think while the plethora of highly specialised, point solution MLOps tools are exciting, maturity demands consolidation and a platform approach is one of the best ways to do that.

Amazon, Google, and Microsoft all continue to lead in enterprise AI with a large portfolio, often through acquisition or through investing in R&D.

A7: It's a large and diverse set of tech and depends on what you want to achieve. Some are good for out of the box regression models for getting started and other are better at more complex uses cases like finding similarities in large sets of data.

@RyRaiker Many would argue that the cloud players will "win" the AI war. But -- there are so many sectors, they can't prevail in all of them.

A7: significant considerations for #AI are cross-domain observability & actionability, parsing out event “noise”, and intelligent alerting & #automation. (2/2)

cloud is a must, but even startups can access and deploy cloud first

Agree, the cloud players provide the infrastructure but not the end-to-end for complex business problems.

@RyRaiker I think this is true. I also think though that analyst results serve more as a north star. Orgs still have to develop the skills to assess their own needs against the features of the product. That means they have to have the in-house talent, or buy in help, to do so