The ScaleUp:AI Conference wrapped up last week. The conference featured speakers from cybersecurity to finance, healthcare, etc., sharing their thoughts on the need for organizations to scale their AI implementations, the way to achieve it, and the top trends in machine learning today.
Artificial intelligence is among the most exciting technologies with the potential to shape an organization’s growth priorities if applied correctly. It isn’t a one-time effort but rather entails continuous and consistent improvement, considering AI is evolving quite rapidly.
“Ultimately, the companies that can’t take full advantage of AI will be sidelined by those that can — as we already see happening in several industries, like auto manufacturing and financial services,†wroteOpens a new window three data analytics experts for the Harvard Business Review.
In fact, 75%Opens a new window of executives surveyed by Accenture agreed that not scaling AI in the next five years may put them out of business. Additionally, 76% said they know how to pilot but struggle to scale AI across businesses.
Aligning the business goals with a sound AI strategy is key to getting the wheels off the ground for organizations in any industry. At the ScaleUp:AI Conference, business executives discussed key differentiators that can help successfully unlock the potential of AI/ML.
Four Takeaways from the ScaleUp:AI Conference
The need for a data-centric approach
To scale AI implementations, the pilot needs to succeed. Andrew Ng, the co-founder of Google Brain and Coursera, pointed out that the failure of companies to make significant returns on investment is the lack of appropriate data that can generate better outcomes.
Ng said a data-centric approach, instead of a model-centric approach that focused on improving the algorithms or code applied to a dataset, needs to take over. This is because algorithms have become as fine-tuned as they ought to be, while organizations lack high-quality datasets because of the amount of relevant and irrelevant data produced today.
As a result, sectors outside consumer tech rarely have access to high-quality data. And even if they do, the data isn’t refined or customized enough for applicability.
Ng’s suggestion? “The world needs a lot more vertical platforms to address issues.†He pointed out how a data-centric approach can cut the amount of time a year-long project may need for completion by as many as 11 months. The solution is to consistently produce high-quality data through all stages of an AI project.
“By addressing small data and customization problems, it is key to democratizing access to AI. Democratizing AI benefits everyone. This community holds the key to unlocking this next era of AI,†Ng said.
See More: How can Businesses Drive True Value and Innovation through Data and AIOpens a new window
Scaling AI
Vittorio Crettella, CIO at Procter & Gamble, also cited the relevance of data and three other factors: talent, platforms and trust, to scale AI successfully.
“AI is embedded in everything we do and embedded increasingly in all our products,†Crettella said, pointing out that his company saves $60 million a year.
First and foremost, the data needs to be usable, irrespective of the source. Companies need to shed the legacy, siloed data architectures. Second, this data should be appropriately leveraged by talent (data scientists, ML engineers, business analysts) to make business decisions.
Third, algorithms should be agnostic of the business segment they are used for, and fourth, bring in a mindset shift and trust the decision-making of AI. Organizations can achieve this by educating employees and relevant stakeholders about how algorithms work.
AI in cybersecurity
SentinelOne COO Nicholas Warner called on the need for autonomous cybersecurity going forward. It basically means equipping cybersecurity operations with AI/ML to monitor and protect organizational systems.
Legacy cybersecurity implementations contributed to the recent surge in successful cyberattacks. What could’ve helped were cloud-based monitoring tools for continuous threat and vulnerability scanning across servers and systems. However, most organizations can’t afford this.
“Having machine-speed decision-making is changing the game,†Warner said. What’s worrisome is that AI is proving to be a game-changer for both businesses and adversaries. He added, “We live in a connected world, which means an enormous opportunity for adversaries.â€
AI/ML is being leveraged to filter out targets and other malicious operations such as the deployment of deepfakes, vulnerability discovery, etc. Warner said, “The future of cybersecurity is AI – not just for defenders but also for attackers. The question remains, whose AI will win − that situation will play out?â€
See More: What Is Machine Learning? Definition, Types, Applications, and Trends for 2022
Machine learning trends
Global head of machine learning at AWS, Allie Miller laid out the trends shaping machine learning.
Low/no code
Aided by the latest generation of language models such as GPT-3, low code or no-code platforms, i.e., platforms relying on visual programming for workflows, data visualizations, designing, etc., are becoming more valuable to develop applications. Low code or no-code platforms are generally targeted at those with limited technical knowledge, meaning machine learning is inching toward adoption by business analysts, besides pro developers.
Decentralization
Decentralization of machine learning essentially means distributing the training of neural networks or ML models across a cluster of machines by partitioning the data and the model. Decentralization has roused interest in recent years to improve performance by eliminating the bottlenecks that arise from consolidated data storage. Research by Travis Addair at Stanford University indicated that while decentralized ML may increase training speed, it can diminish model accuracy.
Security
According to Miller, privacy enforcement is an area wherein organizations leverage ML for compliance and bias mitigation in the use, processing and storage of personal data.
Upskilling and Reskilling
In the last 11 months, Miller said there are now nearly 3x the number of job listings for AI specialists than there are actual job searches. “This is a massive and growing gap and we don’t yet have the skillset to be able to complete that.†Upskilling and reskilling is a way for organizations to build a future-ready workforce.
A recent World Economic Forum (WEF) reportOpens a new window estimated that AI/ML will eliminate 75 million jobs but will also create 133 million new ones. This means the workforce is set for an evolution. The same WEF report also indicated that 54% of all employees would need upskilling and reskilling. According to DeloitteOpens a new window , the U.S-based AI workforce is made up of just 26% women, meaning there’s a huge scope for women to fill in those roles.
Share your thoughts on key developments in AI/ML on LinkedInOpens a new window , TwitterOpens a new window , or FacebookOpens a new window . We would love to hear from you!