Grading Our Machine Learning Predictions for 2017
Grading Our Machine Learning Predictions for 2017
BigML walks us through a grading of the predictions they made last year. Were they accurate? Or did they miss the mark? Come find out!
Join the DZone community and get the full member experience.Join For Free
Some say the easiest way to make a fool of yourself is to try to predict the evolution of technology. As such, making predictions about a field as fast-moving as machine learning is definitely not for the faint of heart. Nevertheless, in our line of business, it’s essential to anticipate the trends that will shape the future impact of machine learning across industries. So, we’ll continue the tradition for 2018. However, to level-set, we’ll start by grading our 2017 predictions and see if our 10 Offbeat Machine Learning Predictions held any water.
Big data soul-searching leads to the gates of machine learning.
The premise was the underreporting of failed big data projects due to technical complexity and the resulting dubious ROI. Although big data as a buzzword has not completely disappeared from planet earth, it certainly has lost its luster in the conference and thought leadership circuit in 2017. However, machine learning and AI remained hot topics of interest throughout the year, pretty much dominating the airwaves and digital channels in the business media. The industry is past the need to label today’s data "big data" much like we as a society are way past calling household electricity "high-voltage alternating current."
No surprises here, so we have gotten this one right!
VCs investing in algorithm-based startups are in for a surprise.
2017 saw continued VC interest in all things machine learning, but for all practical purposes, throwing money at any startup with the mention of the term has been abandoned by now. A new type of wisdom on the role of VC money for AI/ML startups is starting to shape up, which we feel took place more rapidly in favor of machine learning as an enabler than we originally thought. This, no doubt, is a good thing for the longer term viability of the space, which needs to move forward with actual products and real-life applications rather than slide decks and unverified claims of some world-beating algorithm. In general, it’s healthy to remain skeptical of any such algorithm that is not published and peer-reviewed.
The verdict: We may have misjudged the speed of the funding dollars moving away from pure algorithmic outfits, but the fact that there have been no significant exits from such companies in 2017 makes us think we got this trend partially right!
Machine learning talent arbitrage will continue at full speed.
It’s safe to say that the media frenzy has shifted towards Bitcoin and other cryptocurrencies in the second half of the year, giving AI and machine learning a breather. AI/ML wouldn’t be able to match the roller coaster ride of Cagecoin even if they tried, anyway. With that said, the talent hunt for more experienced academics and practitioners in the space is as heated as ever. Nearly all job market predictions for 2018 and beyond show a growing interest in such profiles by the likes of major corporates or their research labs.
So let’s put a check mark next to this prediction: right!
Top-down machine learning initiatives built on PowerPoint slides will end with a whimper.
We’re happy to report that 2017 had its fair share of PowerPoint slides showing a cyborg hand shake hands with that of a human — heartfelt congratulations to the fellow who took that stock photo, by the way — he’s a winner! The suits on stage reciting how many minutes of video is being uploaded to YouTube every minute or which distant star we would have reached by now if we stacked all the hard drives AWS uses in its data centers? Not so much. By now, most top-down efforts have been realized for what they are: dead ends. The industry is prioritizing getting their hands dirty with their own data and investing further into their infrastructure in the hope of turning things around in 2018.
This prediction has fulfilled its mission: right!
Deep learning commercial success stories will be few and far in between.
The deep learning frenzy continued in 2017 unabated. There are more and more courses on the subject and a greater number of open-source tools and packages by the day. Many bright minds still sing praises to its virtues and it continues to dominate research budgets and academic grants. Part of this has to do with the recent advances in AlphaGo like systems conquering more board and video games. Beating the best Go-playing system in quick succession or beating expert human poker players are no easy feats for sure, and they fully deserve the attention they enjoy — even though some experts question whether they represent real breakthroughs.
However, beyond those controlled experiments, there have been few blockbuster use cases for deep learning in the enterprise just yet. This is not to say corporations aren’t doubling down on Deepnets and exploring new areas to apply this somewhat enigmatic approach that still requires lots of trial and error. Maybe 2018 will showcase tangible strides in interpretability issues or unsupervised approaches. Deep learning is here to stay, but 2017 didn’t necessarily register as a banner year for deep learning in the enterprise. It was more of a dip-your-toe-in-and-see-if-the-water-is-warm kind of experience for many companies that aren’t called Google, especially because such systems call for specialized and expensive hardware even if the talent scarcity issue is addressed.
We’ll go with partially right!
Exploration of reasoning and planning under uncertainty will pave the way to new machine learning heights.
Perhaps related to the eminence of deep learning, no serious diffusion was observed in a wide enough section of enterprises when it came to advances and applications in reasoning and planning.
Let’s just say that we were too early with this one: wrong!
Humans will still be central to decision-making despite further machine learning adoption.
Despite all the hoopla about fully autonomous systems and the media’s unbelievable knack to cover news predicting doomsday scenarios with killer robots in charge, we’re still a long way from either being our daily reality. Skilled humans are still heavily involved in every aspect of machine learning systems from data wrangling, exploration, and model training to evaluation, deployment, monitoring, and maintenance. Consider the highly visible example of self-driving cars. We still haven’t been able to make them a reality even though significant strides have been made in the level of autonomy these vehicles have achieved. Full autonomy is a tricky goal, as it has implications beyond the algorithms such as human factors, regulations, economic incentives, and externalities.
This prediction definitely held true in 2017: right!
Agile machine learning will quietly take hold beneath the cacophony of AI marketing speak.
Agile machine learning is not necessarily a buzzword at the moment, but a lot of companies new to the practice have seen the value of starting small with low-hanging predictive use cases in their context instead of launching way-too-ambitious “AI strategies” without legs. Many have already come to the realization that the iterative nature of machine learning projects is better-suited to an experimental approach. Breakthrough moments ultimately take place if one steadfastly pursues the business objectives yet they are seldom delivered in a linear fashion. This situation makes it even more important to a) prioritize the right problems for the business and b) arm more of your subject matter experts with the right analytics capabilities so that they can conduct many experiments in parallel. There certainly are such value lessons learned in 2017, but the Agile approach to ML is not yet the dominant model for all as the wonderment and hype phase wasn’t fully behind us as or year-end 2017.
So, we’ll grade this prediction as partially right!
MLaaS platforms will emerge as the “AI backbone” for enterprise machine learning adoption by legacy companies.
In 2017, more technology companies went on to reveal what makes their internal MLaaS platforms tick and why they’ve invested in them as much as they did. Without a doubt, the common denominator between these companies is their understanding of the links between machine learning and their ability to constantly improve their core products and processes. Hundreds, if not thousands, of legacy businesses with limited development resources have started evaluating cloud ML solutions such as BigML. BigML’s service is now utilized by over 60,000 people and we do get regular inquiries from new users looking to adopt BigML companywide or across adjacent departments.
Still, we realize that we’re in the early stages of this sea change, so we’ll give it a partially right!
Data scientists or not, more developers will introduce machine learning into their companies.
Have you noticed what was common in the developer events for AWS, Microsoft, Google, Salesforce, etc.? Yes, the need to get developers into machine learning! More and more, it feels like the moment we as BigML can interject, “We told you so!” However, we’ll let it pass, as we’re happy to see this trend take hold and become the industry norm as it will go a long way to ease the machine learning talent bottleneck for many of businesses.
We were very much right on this one.
Phew! This covers all our 2017 predictions. Our total score: 7/10. Not too bad for a first shot. Do you agree? Were we too generous with our self-grading? Let us know in the comments or on Twitter.
In the next post, we’ll delve into our top 10 predictions for 2018, so stay tuned!
Published at DZone with permission of Atakan Cetinsoy , DZone MVB. See the original article here.
Opinions expressed by DZone contributors are their own.