Log

A recent Technology Review article citing trends in AI research caught my eye. While we like to say there’s nothing new under the sun, AI research should be literally new–or is it?

The three most common paradigms in the pursuit of true AI are knowledge-based systems, machine learning, and more recently reinforcement learning.

But these aren’t new approaches at all. Many have been around since the 1950s. But as processing speeds increase and neural networking comes into its own, they are all vying for the limelight.

Knowledge based systems were very popular initially because they boiled the knowable world down into rules. If there’s a thought, there’s a rule for it, the logic went. So researchers built more and more rules until the systems were such a glut of rules that they couldn’t get out of their own way.

Supervised learning, using neural networks, got a big boost in 2012 when Geoffrey Hinton and his team from the University of Toronto beat everyone at ImageNet by more than ten percentage points–moving the needle not incrementally, but massively. But deep learning requires so much labeled data that it can take weeks and require petabytes of data to build a knowledge set with discriminative capabilities. This is what we generally think of as Machine Learning.

The new, old, kid on the block, Reinforcement Learning, has seen a huge increase in research paper mentions. Reinforcement “mimics the process of training animals through punishments and rewards.” Reinforcement languished for many years until 2015 when DeepMind’s AlphaGo beat the Go world champion.

So whether its Reinforcement Learning, Knowledge Based systems or Bayesian networks, no one knows. In the article Pedro Domingos of the University of Washington and author of The Master Algorithm says either another older paradigm will rise to the surface again or we could see something entirely new.

I am one of those blithe technologists who doesn’t really care that Google knows where I am, what I’m interested in, and who I know. But I also know that I don’t like Facebook for exactly the same reasons. As I am a pastiche of the choices I make, I accept some cognitive dissonance related to something like data.

And if you insist on either total privacy or data freedom, then I would ask you if you only eat foods that are healthy for you or bad for you? Do you exercise every day or when you can, or just when you feel like it? Do you never speed when driving, or do you speed sometimes?

Just as with life choices, we make decisions about our data the same way. This is both good and bad. When a birth date is demanded of me by some website, I often lie. Their data is now corrupt, but usually within a few days of my actual birth date, so for their purposes perhaps close enough is good enough.

Ideally, our data would flow anonymously to all the corporations and governments that need to make use of it. Right now, of course, the US Government doesn’t collect a lot of data that could protect us, and does collect some stuff that constitutes an invasion of privacy.

Of course Big Data can (and will) be used for troubling purposes. But it could also rectify current injustices caused by ‘broad brush’ data mining. If you have perfect credit but live in a neighborhood and have an income that would generally suggest you aren’t credit-worthy, broad brush data might suggest you not receive a mortgage to buy a house.

In other words, I don’t believe we need to know all the ramifications of varying levels of privacy or lack thereof before moving forward into a new, autonomous age. Humans are really good at solving problems. They also create a lot of them on the way.

I recently came across a new podcast over at QZ, addressing the issue of whether all technology is good all the time, called Should This Exist?

Questions about neurohacking, CRISPR babies, empathetic robots, etc. abound, but it’s obvious, a lot of tech is not good at all.

Neurohacking, where you either use drugs or electricity to stimulate the laying down of memory pathways in the brain might be great, right? But since we just barely learned how the brain works, and not even that, entirely, I’d rather stick to practicing my skills and using something like Neuro Linguistic Programming (NLP – confusingly the same acronym as Natural Language Processing) to get me there. And it was scientists who told us that SSRIs would solve all our emotional problems, not understanding that removing the SSRIs would result in neurochemical imbalances in the brain for, possibly, the rest of the user’s life.

CRISPR could be great at saving lives and eliminating genetic disease, but could also be used for what racist scientists in the 19th and 20th centuries called Eugenics. I don’t know if humans are yet wise enough to wield such powerful technology.

And of course, Big Data looms over all. A lot of people don’t want their data out there. A lot of people believe that our data will be misused by corporations, governments and law enforcement. Benjamin Franklin said, “Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety.”

I don’t know what Liberty means today, and I’m not sure what Safety is, either. Is Liberty being able to do things without anyone’s knowledge? Or is Liberty being free do what we want. I’m a big fan of Benjamin Franklin, but I’m an even bigger fan of Franklin Delano Roosevelt. His Four Freedoms speech defined what we, as society, should be aiming for for all our citizens.

And if we’re looking for universal Freedom of Speech, Freedom of Worship, Freedom from Want and Freedom from Fear, technology supports those things. Technology has enabled so much Free Speech right now that it’s actually a burden to call out idiots on their racism and ignorance. Even if you live in deepest, darkest Mississippi, you can choose your faith, and you can be an active online member of any religious community. We haven’t reached Freedom from Want, but technology, as always, could get us there–the main block to advances on the Want front come from greedy, frightened humans. Lastly, how do we free ourselves from Fear? Education and knowledge are the only medicines for that, and technology is bringing (along with all that mucky free speech) more information to more people than ever before.

And it can be scary that the G-MAFIA tracks almost all our public data. But when one of them went rogue, selling our private data to Russians (Facebook), a lot of people called them out. Were they pilloried? Pretty much, yes. But they weren’t forced out of business. We like their business, and the Russians and Cambridge Analytica paid them to buy our publicly shared information. I don’t know, but if you’re shocked by a business selling what you give them for free, you really need to bone up on your privacy practices.

In the end, things that exist exist, and those pesky, curious humans will not stop inventing new, ever more dangerous tools. You can’t put the genie back in the bottle. Far better to try to process our anxieties and agree through the social contract what appropriate protocols should be put in place.

Some really exciting stuff is happening in the natural language processing (NLP) field. If you didn’t already know about it, Google Duplex, the successor, or upgrade to Google Assistant, has already rolled out in key markets (if you don’t live in NYC or San Francisco you probably don’t have access).

OpenAI researchers have created a process by which AI “learns” from unlabeled text. This is essentially the process of letting the AI teach itself language by picking it up on the fly, the way humans learn it. This is a major step toward real AI conversations, where the AI isn’t just interpreting keywords related to a set menu.

Another mind expanding program at Google is developing a structure for AIs to fill in words it can’t hear.

Once these roll out, your Google Assistant will become an RPA all its own. This will revolutionize your personal life. Why not find out how RPAs can help your business in the same way? Check out real-life interactions between Google Duplex and humans here.

No real surprises in a February 22 Information Age article by Andrew Ross discussing issues relating to RPA implementation, but he does point to the biggest issue facing all autonomous technology uptake: understanding what needs to get done and how.

As “swivel chair” software, Robotic Process Automation (RPA) promises to be a game changer for small to midsize businesses looking to increase worker productivity by reducing the amount of time they spend doing human-as-machine labor. But many workflow systems have developed organically, over time. This organic growth incorporates a lot of activity that may no longer make sense. Automating the process as-is can build in inefficiencies.

But Ross’s article illuminates the big problem facing business owners: Alex Rinke, CEO of Celonis, told Information Age he’s “heard of many failed RPA initiatives; and it’s often because enterprises fail to understand their inefficiencies before implementing automation.”

There’s no doubt about it, RPAs and Machine Learning are going to radically improve the speed of your business. But you need to work closely with system specialists like ATC to understand what needs to get done and how best to do it.

NYU professor Amy Webb is coming out with a new book on March 5th, The Big Nine: How the Tech Titans and Their Thinking Machines Could Warp Humanity.

The central tenet of the book is that AI is currently being developed by commercial enterprises (the G-MAFIA–Google, Microsoft, Amazon, Facebook, IBM and Apple) in the US and Tencent, Alibaba and Baidu in China, while the US government has decided to turn its back on research funding, and Europe is figuring out the web.

My favorite anecdote from the congressional Facebook hearings is a septuagenarian senator demanding of Mark Zuckerberg, “How do you make money with this thing?” To which a stunned Zuckerberg replied, “…We sell advertising, senator.”

This laissez faire attitude on the part of governments is a mistake, Webb, asserts, because these businesses are only interested (as they should be) in turning these technologies toward profit.

She suggests an international guiding AI body be created, GAIA (Global Alliance on Intelligence Augmentation), to ensure societal benefit from the implementation of AI systems. AI is a wave that has not even created a visible swell on the horizon. It would be far better for governments and societies to be prepared than to let that wave hit the beach at full force without any preparation.

As AI is only currently applied to rule oriented data systems, it feels distant. But as the learning sets grow, and applications increase in complexity, we will see more and more life quality systems subsumed by this nascent wave.

In the recent O’Rielly whitepaper, “The Path to Predictive Analytics and Machine Learning,” the authors , Conor Doherty, Steven Camina, Kevin White and Gary Orenstein, point out a key issue facing many businesses: data silos.

What is a data silo?

Traditional data architectures use a siloed, Online Transaction Processing (OLTP) model for Customer Relations Management (sales, returns, queries), and a completely separate data store for analysis (if these are accessible online, they are called Online Analytical Processing (OLAP) warehouses.

OLAP-optimized data warehouses cannot handle one-off inserts and updates. Instead, data must be organized and loaded all at once —as a large batch—which results in an offline operation that runs overnight or during off-hours. The tradeoff with this approach is that streaming data cannot be queried by the analytical database until a batch load runs. With such an architecture, standing up a real-time application or enabling analyst to query your freshest dataset cannot be achieved.

Doherty Camina, White, Orenstein

The O’Rielly team suggest partnering the somewhat alarmingly named Apache Kafka, a high-throughput, distributed messaging system that “acts as a broker between producers (processes that publish their records to a topic) and consumers (processes that subscribe to one or more topics)” with Apache Spark, a distributed, memory-optimized system for data transformation.

envision a shipping network in which the schedules and routes are determined programmatically by using predictive models. The models might take weather and traffic data and combine them with past shipping logs to predict the time and route that will result in the most efficient delivery. In this case, day-to-day operations are contingent on the results of analytic predictive models. This kind of on-the-fly automated optimization is not possible when transactions and analytics happen in separate siloes.

Doherty, Camina, White, Orenstein

While your company might not be ready for real time analytics, its worth thinking about building a system ready for transformation. And good news for your bottom line, Apache Kafka and Spark are both free.

We hear a lot about Big Data. What people don’t talk about much is that storage of all that data is going to be extremely costly, both financially and ecologically.

But nature, from the beginning, developed the most efficient memory storage medium in existence. The DNA of one human, as Riccardo Sabatini dramatically presented in a 2016 TED talk, fits into 175 printed volumes that are each 1,600 pages long, with a font size of 6 (1/2 of what you’re reading). He also pointed out that the information representing every molecular aspect of a newborn baby would fit on 2000 Titanics packed with thumb drives.

In 2012 two Harvard scientists, George Church and Sri Kosuri, figured out how to translate data into DNA. Since then, DARPA has granted a lot of money to companies and university researchers to figure out how to make DNA data storage a reality.

Why? DNA is shelf-stable. It can last, as we know from Woolly Mammoth samples, 60,000 years. DNA is not electrically or magnetically activated. DNA is also ridiculously dense. To illustrate: every film ever made could be saved on a pile of DNA the size of a sugar cube.

To take it even further, every piece of data currently in existence, stored on DNA, would fit in a single, average sized closet, requiring no fans, cooling systems, electricity or hardware. It would never degrade.

Right now, a company called Catalog is building a concept machine that is the Guttenberg printing press of DNA memory storage. The size of a school-bus, the encoder/decoder processes base pairs like movable type. Instead of the insanely costly early methods, Catalog uses shorthand to encode binary sequences.

If you hired Catalog to put every photo, movie, piece of writing, homework, report card, textbook, tv show, novel, social media post and medical record you’d ever generated or consumed in your life–basically everything that can be recorded–into a bunch of strands of DNA, it would fit in a drop of water. And it would last basically forever.

Google and Microsoft are both researching DNA storage. We need it now.

Brian Bergstein, in the article “This is why AI has yet to Reshape Most Businesses” in MIT Technology Review, comes to the conclusion that most companies don’t have the time or resources for enacting AI programs.

AI is expensive, and unless managers can see a real need for it, it will remain a technology used only by the most profitable and data driven industries.

But that doesn’t mean that you should be complacent about other autonomous technologies. Achim Daub, a perfume executive says, “no one really has time to do greenfield learning on the side.” He was speaking about AI development, but I couldn’t disagree more. It is precisely ‘greenfield’ exploratory learning from which your company’s biggest transformation will arise.

If your VP of Sales was able to compile reports in seconds rather than days, would that transform your business? Of course it would. Robotic Process Automation is a relatively new technology that most managers haven’t had time to learn about. Set your business’ managers free to learn what RPAs, robotics and machine learning could do for your company. Let ATC be their guide.

When I was working with McKinsey & Co., the most repeated piece of information in efficiency studies was that people can only really work productively in an office for about three hours each day. The rest of the time is spent doing tasks that don’t require human-as-human labor. As those tasks shrink–consider the assistant whose day used to be spent retyping, copying and paper filing–what would you like your employees to be doing with their time?

Humans are much better than computers at learning, and learning is the one redirect that can effectively refocus your employees once their initial “three hours” is up. At ATC, we suggest at least two hours per week per employee of professional development time to learn new software, keep up-to-date on industry innovation and improve skills. But we also suggest three hours per week of exploratory learning.

Were you to include exploratory learning as part of your office workers’ job requirements, what would that look like, what would it do, and what metric could you use to measure it?

“Old dog” employees, those who just want to run down the clock to retirement, are least likely to take advantage of exploratory learning. Also, unmotivated “users” will see exploratory learning as an opportunity to goof off. They see their employment as a transactional scam anyway. You take their time and they take your money, and they are always interested in giving up less time and getting more money.

But employees who believe in your company’s mission, who are able to envision growth in their job and eventual promotion, will use every second of exploratory learning becoming a better employee.

Exploratory learning is just that. It is not professional development, where, say, someone in accounting takes an IRS information course. It is when someone in accounting decides to learn 3D modelling or takes violin lessons.

What could this possibly do for your company? Humans have been building tools since the dawn of time. But the stick you use to hit an enemy is also useful as a lever, a walking aid and a probe. 3D modelling, for example, triggers connections between the concrete and conceptual structuring capacities of the brain. Musical notation is a value coordinate system of unsurpassed nuance. Study in either of these fields can lead to important breakthroughs in approaching problem solving in the office. Also, by telling employees that you want them to learn new tricks, they know you are investing in them. In very real terms, this leads to greater employee loyalty.

How do you measure it? Before implementing any digital transformation, ATC would recommend you take a baseline survey of your employees querying everything from productivity and workload estimates to happiness and quality of life evaluations. Once you have that baseline, introduce exploratory learning and regular professional development and see how they feel in six months.

We think you won’t have to take a survey. You’ll see the increased productivity on your P&L report.