I was recently interviewed by #TresconTechReview ahead of a talk that I gave at the World AI Show — Qatar, 2020, and this is the transcript of the interview (with some edits). In this article, I’m going through some market trends and observations on the latest developments in the world of AI and Analytics. Hope you enjoy the reading. Feel free to comment, especially if you see things differently or simply have a different perspective!
1. Thank you for taking out time to chat with us. To begin with, how do you see the landscape of AI and data analytics developing globally? What specific trends do you see that will shape up the future of its adoption?
Thanks for inviting me!
In recent years, the AI landscape has seen a dramatic acceleration of the level of adoption, globally. This is due to a combination of factors starting with some significant improvements in technology areas such as neural networks, deep learning, image recognition and natural language processing. In parallel, data processing has become commoditized through open-source and cloud technologies.
Organizations are also moving towards XaaS (Everything as a Service) consumption models so that they can innovate faster without having to massively invest in hardware, software and services. The world where organizations buy a software, then spend tons of money on hardware to install it and then months to configure their solutions, is over. The technology has become easily accessible, cheap, and innovation is achieved through short, agile iterations allowing to fail fast and try often. Analytics users want instant access to the latest versions of enterprise analytics software, they do not want to worry about upgrades, and they want to pay for what they use, or for the results they get. In effect, self-service analytics is fuelling the democratization of AI.
This evolution of the technology landscape and practices is making it possible and relatively easy for private and public organizations to implement innovative use cases and to leverage AI to build greater competitive advantage and develop new business models.
On one side, this is great because AI is proving to be a catalyst for digital transformation, and it creates many new business opportunities.
On the other side, organizations are now facing an increasingly complex ecosystem that is difficult to navigate. New technologies appear every day, with no readily available recipe. Lots of choices have to be made in order to decide what pieces are relevant and how they fit together. So, what was supposed to make it easy to access and adopt those technologies is having the opposite effect of making it difficult and not delivering the expected results.
As technologies are becoming more accessible, it is relatively easy to develop sophisticated models and to experiment with new use cases, but the real challenge lies in the operationalization of those models, embedding analytical insight into the decisioning fabric, at the front line of business operations.
The danger is for organizations to go down this ‘rabbit hole’ of never-ending experimentation, lack of business results, and increased technical debt.
Finally, growing awareness in the general population about the applications of AI technologies, as well as regulatory pressure, are raising the need for more governance, transparency and explainability in the way those technologies are used, how automated decisions are made, and the impact they have on people. I believe this question of the ethics of AI will have a significant impact on the adoption of AI technologies and their applications in the real world. This is likely to hit the headlines in 2021 as the European Union is cooking a new regulation for the responsible use of AI, along the same lines as the GDPR, and this will most likely have a ripple effect across the world.
2. In a pandemic like Covid-19, what are the most important recent developments in the field of data and analytics and how do you believe it will impact the industry?
One thing is for sure in this crisis: there is no shortage of either questions to answer or data to analyze. Uncertainty and fear are driving this appetite for answers, but it is hugely encouraging to see that data and analytics are helping us to answer the call.
It started with the need for governments to use data visualization techniques to inform citizens, to make sense of the unfolding events, and to inform policies designed to slow down the propagation of the virus and minimize the impact on communities. Very quickly, individuals with access to relevant data or with the right skills have joined the collective effort to fight the pandemic. Countless crowd-sourcing initiatives have been launched. Academic institutions have also mobilized to help governments navigate the crisis and inform their decisions and policies. Tech companies rallied to do their bit, helping customers and communities through the storm.
Ethical Robots Are a Naive Illusion for Programmers? | Data Driven Investor
The relations between rhetoric and ethics are disturbing: the ease with which language can be twisted is worrisome, and…
What this pandemic highlighted is not only how data and analytics can be used to navigate a crisis, but also how much we rely on data to make sense of unforeseen events and to adapt to survive in fast-moving market conditions. Many organizations realized that they needed to build more resilience in their operations, finances and infrastructure.
Analytics plays a big part in strengthening this resilience. Organizations should be able to rely on their data and analytics capabilities when they need it the most. Data should be readily available and reliable so it can be used to fuel decision making. Self-service data visualization and exploration capabilities can help to quickly extract hidden insight from the data without having to call on specialized technical skills. Predictive analytics and decisioning capabilities are critical instruments for Captains to steer their ships in the storm and beyond. How successful they will be, depends on how quickly they can access the data they need and how effectively they can use this data to make informed and timely decisions that will save lives, jobs, and businesses.
Effectively, the crisis is proving to be a catalyst for change in business practices, and for the adoption of a more data-driven decision-making approach. I believe that this momentum will be sustained over time and amplified by the quest for greater resilience.
3. What in your opinion are the most pressing challenges that inhibit AI implementation ? What can be done to overcome them?
Today, the biggest barrier towards successful AI adoption is the difficulty that most organizations are facing in operationalizing their analytics, i.e. efficiently moving analytical models from the innovation lab into the real world where they can drive business decisions and impact the bottom line.
The deployment of models in a production environment requires a number of ingredients:
- Effective collaboration between business analysts, data scientists, application engineers and IT operations.
- The integration of multiple environments and the automation of the analytic lifecycle through CI/CD/CM (Continuous Integration, Deployment and Monitoring) techniques.
- A governance framework allowing for a value-driven orchestration of analytical efforts, a centralized management of all analytical assets, and things like transparency, traceability, and explainability to tackle the challenge of algorithmic bias.
Ineffective processes lead to excessive time to value and excessive efforts spent in the building, deploying and monitoring of analytical models. With the pressure on cost management, analytics teams are expected to deliver more with the same resources.
Other challenges include the difficulty to find and retain data scientists’ talents, and the lack of clarity on the expected value that AI projects should deliver.
How organizations can overcome those challenges is by implementing a ModelOps framework and using an open, cloud-native platform approach to orchestrate and simplify the analytics ecosystem.
4. How do you think Cloud Computing is going to shape the future of AI and Analytics?
The question of migrating Analytics workload to the Cloud is a matter of WHEN and HOW rather than IF!
It’s basically a “no-brainer” and most organizations have already taken some steps to experiment with Private and/or Public Cloud Platforms in order to achieve the agility, flexibility, and reduced TCO they’re looking for for their analytics.
There are clear benefits that can be achieved by running analytics in the Cloud:
- Proximity to the data needed to score models and to the front line applications using analytical insight to drive real-time decisions.
- Well suited to support CI/CD/CM techniques, bringing the automation and repeatability needed to scale analytics and deliver fast time-to-value.
- Natural fit with the move towards usage-based consumption models.
- Readily accessible analytics capabilities without the need to invest in hardware and to install software, supporting the need of citizen data scientists with code-free approaches.
However, in my view, it is still very early stages, and most early adopters have not yet achieved the expected benefits because they have been unable to move from the experimentation phase to a fully operational production environments.
My recommendation is clearly to embrace the experimentation phase in the “Lab” because it drives the creativity of the data science teams, and it helps to identify and prioritize use cases. However, it is equally important to also think through a proper strategy for cloud analytics with a clear end-goal in mind, because the choices made in the Lab are not necessarily the best when it comes to scaling and governing analytics in production.
We need to pay attention to things like:
- Data gravity: In the world of cloud computing moving data is expensive. Do we have the appropriate data in the Cloud to train, score and monitor models?
- Multi-Cloud strategy: How can we make sure that our architecture and technical choices are not tied to a specific cloud vendor? How easy would it be to migrate from one cloud platform to another? How can we manage and govern data and analytical assets across multiple platforms?
- Technical debt: By introducing a lot of new components and tools across hybrid cloud and on-premise environments, the risk is to aggravate the complexity of the analytics ecosystem and to worsen the challenge to find (and retain) the appropriate skills to manage those environments.
In summary, Cloud Analytics is a key enabler for digital transformations, allowing organizations to innovate with agility. However, organizations should establish a clear strategy in parallel of experimenting with those new techniques and tools.