eWeekChat

   4 years ago
#eWeekChatMulticloud ChallengesExperts discuss multicloud computing
   4 years ago
#eWeekChatFuture of Edge Computing Experts discuss the Future of Edge computing.
RFPIO
A4: Starting at the top. Leadership has to fully commit to AI and see this as an enterprise change management challenge and bring along the broader organization. Currently it's the domain of a few in R&D labs.
Jens Doerpmund
A4. Companies need to develop standards and tools for identifying privacy violations and bias, ensure accuracy and provide explainability. Data quality and governance will remain relevant
BMC Software
Explainability is indeed a huge issue, especially for adoption. That’s why we had to build a knowledge graph based AI architecture to reason about causality as opposed to closed, non-attributable pure ANNs.

(edited)

James Maguire
Q4. We've talked about this somewhat: How do you recommend addressing these AI challenges?
Brian podolak
Q4 well let's chat about bias, in reality, the everyday data the organizations collect is poor and holds no significance of its own.
Bonnie Holub
A4. When addressing AI challenges, a disciplined approach is essential. Early in the project it is essential from a perspective of getting the right data sets together, doing the data wrangling, and being able to build the models with the associated exploratory Data
BMC Software
A4: Focused vendors with data advantage will dominate their corresponding verticals.
Bonnie Holub
A4 Analytics...

(edited)

Brian podolak
@BMCSoftware Good point - Data Scarcity!
Bonnie Holub
A4. o Having a process where you are doing this in a disciplined and defined manner is uncommon but it is a differentiator for us.
Trent Fierro
A4. It takes time, but a before & after visual helps. From an org’s perspective, seeing how problems are solved more quickly or that a help desk is taking fewer calls is key. The ability to measure the effectiveness of #AI (or anything for that matter) can’t be beat
Bonnie Holub
A4. o Once you start scoping up, then the data engineers become even more important especially when you have demanding real time needs, audacious throughput, and the dynamic scale up and down of large parallel processing environments as most of our clients do.
Signifyd
@Trentf_CA Definitely agree--being able to measure efficacy and results is vital.
Philip Cooper
@bonnieholub often an afterthought.....IMHO analytics need to go hand in hand with the ML
Philip Cooper
A3. Trust. Explainability. Bias.
Philip Cooper
These are related and actually get harder with generative and LLMs.
Signifyd
A3. A specific problem driving the development of a solution can make the difference between a middling result and an excellent one. (2/2)
Jens Doerpmund
A3 Cost is only one of the problems. Keeping up to date with the latest innovations may be an even bigger problem, not to mention ethics, bias, transparency…
Philip Cooper
agree on your latter point. Interesting research happening in the XAI space but nailing the end-user experience for trust is hard
BMC Software
Bit-rot is indeed a concern. We believe a hierarchical training approach helps. That’s why we tiered our LLM with additional ANN layers, so that we can train HelixGPT with tribal knowledge, without having to teach it the basics of human language generation.
Bonnie Holub
A3.
That’s why we focus on them early and repeatedly through our projects. It’s the differentiator to getting our projects into production and generating value for our clients.
Signifyd
A3. One of the biggest challenges is knowing how to choose solutions that will give you a good return on investment. Some companies have #AI solutions in search of problems, and unsatisfactory results. (1/2)
Trent Fierro
Unfortunately ppl are spending time playing with #AI, so it doesn't help overall...
Philip Cooper
@trentf_CA I think we've seen a shift from experimentation in the lab to use in the front and middle office. Evidence: there’s increasing demand for software engineers to integrate and deploy AI-powered experiences and less pressing demand for data science resources.
RFPIO
A3: Having clear vision for adopting AI across the enterprise, and translating it to a cohesive implementation plan continue to be the biggest challenges. Cost will become a challenge if the strategy and plan are flawed leading to lack of ROI and cost overrun.
Philip Cooper
100% agree....if you skip this, good luck
Bonnie Holub
A3.
Cost is not the big problem, value realization is the big problem. Figuring out the long-term revenue potential or the near-term cost savings associated with implementations is the problem.
Philip Cooper
agreed on cost. This may change in the medium-term if generative AI is adopted in the enterprise