There’s No AI Without IA
There’s No AI Without IA
So you’re ready to leap into developing an AI strategy for your organization. You’re not alone. Thanks to obvious ROI from implementing AI powered solutions, as discussed in my blog Automation + Augmentation Equals Intelligent Transformation. An increasing number of companies are asking the same critical question: where do we start?
First, let’s get a common misunderstanding out of the way: AI isn’t magic. Even current state-of-the-art applications can be exceedingly narrow, in that they deal with structured, organized data to solve a finely delineated problem. A common trap that executives fall into is to overestimate the capability of AI, without realizing how much data a single effective algorithm requires, and — more importantly — the type of data organization that underlies the most robust and flexible AI for your particular application. As the adage goes, “garbage in, garbage out.” Throwing heaps of loose digital files at an off-the-shelf algorithm will not benefit your productivity or profits.
This does not mean, however, that AI implementation should be relegated to only a small group of highly-specialized experts within the company. It shouldn’t. Rather, decision-makers should understand what aspects of their enterprise can benefit from adopting AI, both within and across departments. The process, according to Rob Thomas at IBM, is akin to climbing a ladder.
Scaling the AI ladder
To effectively outline an AI vision, executives need to grasp a basic principle: current AI approaches critically rely on organized datasets and an efficient data architecture. The best place to initiate your AI journey is to identify a particular business problem for which tons of organized data is available, and build an integrated infrastructure to realize its potential.
Why? In essence, implementing AI is like scaling a ladder. Let me explain.
Rather than climbing up, let’s head down that ladder, starting from the top step, AI, to get to the crux. Most robust AI applications these days apply machine learning (ML), a group of greedy algorithms that perform better when given an adequate amount of training data examples. Note, I said “adequate,” not “enormous” or some other variant — more isn’t always better.
ML isn’t magic either. It relies on analytics and statistics to extract patterns and trends beyond human capabilities. Analytics, in turn, critically depends on clean data organized in a way that machines can easily digest. And here’s the bottom rung of the ladder: the dataflow needs to be built in tiers that effectively meets the need of ML, and thereby AI.
In other words, AI begins with information architecture (IA). Or as Seth Earley famously puts it in IEEE Software, “there’s no AI without IA.” I highly encourage you to give the article a read: it not only describes how AI will impact our lives, but it also dives deep into its different applications, decision factors, governance and — of course — the importance of developing strategy and structures to manage data for AI.
In other words, AI begins with information architecture (IA).
What is IA Anyways?
Fundamentally, IA is just a way to organize and structure information within a product or service, so that it can be easily accessed to meet the needs of whoever is querying — say, humans and machine learning algorithms.
There’s nothing mysterious about IA. Some even consider it common sense.
The ancient Egyptians adopted a rudimentary version when organizing their vast libraries to make books more accessible. With the boom of the internet and data explosion, however, IA grew into a vast technical field. Rather than a single idea, IA is now a broad container that encompasses many thoughts and deliverables in smoothly handing queried data to the user.
When optimally implemented, users don’t even notice its existence. These days, IA works in the background of websites, music libraries and photo apps — among other services — to organize web content, blog posts, music and photos files according to tags and metadata. IA is the driving force behind UX design, allowing content providers to deliver accessible, clear, usable, valuable, and sharable content right to our noses.
Ok, so what does that have to do with AI?
IA supports clean data
IA feeds into AI in two inseparable, intertwined ways:
· First, it provides structured data to the algorithm;
· Second, it allows that data to be processed in a broader logical hierarchy to better extract meaning patterns, results and conclusions.
Let’s talk about the first point. Here’s where we’ll dive, just a little bit, into the high-level mechanics of ML, the powerhouses behind AI applications. ML itself encompasses many flavors of algorithms that extract patterns within data, without the programmer explicitly telling them what to do. Regardless of their particular flavor, however, all ML applications require structured data. This is probably easier to grasp in applications that deal with financial and transactional data. Where it becomes a head scratcher are more “clever” applications such as computer vision, natural language processing and other ML functions that deal with images, visuals and language. Unlike spreadsheets full of numbers, this type of information generally doesn’t have a predefined structure or overall architecture to analyze.
Nevertheless, for ML to adequately process this data, they still require parameters associated with the source and context. As one example, an algorithm that analyzes the effectiveness of a marketing campaign may monitor its customers’ social media feeds and perform sentiment analysis. Although the written language can be all over the place, the posts still need to be organized by various parameters that describe their users, posts, relationships, links, hashtags and so on. Here, the first step is to characterize the structure of input data, so that the algorithm can then be passed on to further tiers to identify patterns of interest — for example, the users tend to respond more positively towards the brand following a particular outreach effort.
IA supports efficient data processing
The above example already touches upon the second, and perhaps main, point of IA: an effective processing hierarchy.
To pull back from structured data to data structure, we can delve into another use case– conversational architecture underlying marketing and sales virtual assistants.
Similar to sentiment analysis, here the AI needs to find patterns from unstructured data — the customer’s chat history — to better assist his or her needs. The first step is to develop conversational intelligence, in which all sorts of previous conversational data from across modalities — emails, phone calls, marketing materials — are organized to fed into the AI.
Next, the algorithm identifies patterns from that data, including intent, topics, tags, and (yes) sentiment, which are shuttled into the conversation architecture that directly maps to what customers are looking for.
From there, the next tier is to deploy optimized conversations across teams, including marketing, sales and support. The flow of data is highly organized from intake to output, which streamlines the application, and all teams are set to benefit from adopting AI.
In a nutshell, although ML is particularly efficient at finding structure from unstructured information, at the data layer it still requires organization. As Early succinctly summarizes, “clean data is the price of admission” for ML. Before ever fed into an ML, the data needs defined attributes (or variables), normalization and cleansing. From there, the data needs to be processed in a logical, efficient architecture to better address your final needs, goals and, of course, the company’s bottom line.
Is Implementing IA hard?
Sure, there are stumbling blocks, and there is no one size fits all when it comes to IA for any particular company or enterprise. One common problem is that multiple departments may not require the same type of data or same level of data integrity and organization. This could result in fractured efforts at architecting information to meet each team’s own particular needs, without consideration of the broader company or longer ambitions. An even thornier problem is if teams did not initially decide on a common nomenclature, definitions and other parameters to organize shared “core” data, or when communication breaks down between data science experts and project managers.
So the first, and perhaps hardest, step in building a solid IA foundation isn’t technical. It’s cultural: establish teams that can marry the rich business experience of managers with the technical know-how of data science experts, while educating employees on the value of maintaining clean, organized datasets.
Low code solutions
Now let’s talk technical.
Coding IA from scratch is no doubt difficult. The good news is you don’t have to. Thanks to services provided by AI giants including Microsoft, Amazon, Google and IBM, there are now consumer-friendly IA solutions that even startups and smaller companies can easily adopt and utilize without having to worry about the nooks and crannies of initializing and maintaining the data infrastructure.
These packaged solutions are often referred to as low-code/no-code software, and they’re close to magic. They have the super-ability to empower staff in smaller companies to become citizen developers who are capable of creating robust applications tailored to their own needs. In essence, this type of software provides a user-friendly, visual integrated development environments (IDE). Rather than dealing with code, a user can drag-and-drop application components like digital Lego bricks to link together into applications. The end user experience is like moving from DOS to Windows.
Because they are battle-tested and backed by tech industry’s biggest players, no-code/low-code systems provide a safe, scalable platform that offers a variety of tools to ensure security, updating, testing and sign-off processes with minimal in-house IT support. As a result, the development platforms lower the cost of entry, both in terms of hardware setup and personnel training and deployment.
One such solution is Microsoft’s Common Data Service (CDS), which not only securely stores data but also forms a model with pre-built relationships and rules regarding dataflow. This releases the user from having to worry about enforcing those rules or coding them from scratch. In early June of 2019, the company further developed the AI builder. The builder intakes data in the CDS for analysis, acts on those results through applications and further automates business processes. So far, the low code software suite supports binary classification, simple object detection, business card reading, form processing, and more applications that will likely be added as the system further matures.
Both CDS and AI builder are integrated within Microsoft’s larger Azure cloud platform, which leverages Azure Machine Learning used to build, update and maintain applications (here’s a nifty explainer video). Rather than data scientists hand coding projects from scratch, applications can now be efficiently built by automating certain (simple and tedious) aspects of machine learning, or tougher tasks normally relegated to ML specialists. As an example of the latter, Azure can help you select the most appropriate ML model for a particular scenario and dataset, a key first step in machine learning.
That’s huge. As Microsoft explains: “Traditional machine learning model development is resource-intensive requiring both significant domain knowledge and time to produce and compare dozens of models.” The platform isn’t theoretical. It’s already taken off: prominent examples include smart recommender systems for online marketing or customer service, or — as we have successfully built ourselves — automation and workflow configuration tools to galvanize workplace efficiency.
To sum up, these integrated IA solutions allow executives to focus on the bigger picture of their core business, rather than the nuts and bolts — encryption, deployment, disaster recovery, etc. — of managing a unified data structure. Without doubt, no-code/low-code solutions are already helping expedite companies interested in adopting AI and automation as their core drive, and we foresee this trend gaining further traction.
Social27 Automation Studio
My company has also developed a no-code solution called the Social27 Automation Studio. We utilize it to build conversational automation solutions — AI assistants for chat, voice and content recommendation engines, to name a few.
The take-home message
So. That was a lot. Here’s the takeaway.
Successfully implementing AI to achieve extended intelligence requires buy-in at both the cultural and technical level of a company. For the latter, this begins with organized data and an adequate data infrastructure.
Rest assured, it’s not an uphill battle: although IA is fundamental to AI, it doesn’t require a hefty wallet or years to implement. Off-the-shelf solutions — often called no-code or low-code, are readily available, and they’re increasingly integrated into final use cases. Although there still isn’t a button to automatically generate ML algorithms tailored to your needs, the roadmap towards adopting business AI is becoming increasingly simple.
For higher-level management, perhaps a more pertinent question is that of strategic foresight: what problems warrant the trouble and investment in AI to best reap its benefits? We’ll take a deep dive into some use cases next blog.
Partnering in your AI automation projects
At Social27, we strive to understand your business needs and jointly develop the right AI automation strategy with the help of a step-by-step framework and best practices from across the industry. Subsequently, our automation products and solutions can enable rapid deployment and value realization.
Please visit social27.com for more information. We would appreciate the opportunity to partner in your intelligent automation journey.
Social27 is an AI powered Automation, Augmentation and Analytics platform. Social27 Playlists empower B2B marketers to deliver a personalized and ‘binge-worthy’ content experience in real-time. Our AI Assistant identifies buyer intent, qualifies leads, and schedules meetings; accelerating sales and revenue growth. Social27 Deal Room automates all the repetitive and time-consuming processes in the deal workflow, keeping all calls, contracts, and compliance in one place. And finally, the Partner Ecosystem Accelerator enables velocity across your ecosystem by enabling a #NoFriction customer journey.