Everywhere you go these days, it seems like someone is talking about artificial intelligence (AI). It is the story on every magazine cover and the keynote topic at every conference. In a recent study of over 1,600 respondents, including developers and data scientists, 61% stated that their businesses are making AI their top data priority in 2019, regardless of company size. In the same survey, 65% of respondents indicated that they aim to use AI to make better business decisions, with the majority believing the technology has the potential to transform their industries.
All of the hype is not without good reason. Well-implemented investments in AI are delivering significant returns for companies across a wide range of industries. From Blue River’s LettuceBot that is revolutionizing agriculture by reducing pesticide and fertilizer use to ChemPlanner that aides organic chemists in finding novel synthetic pathways to speed drug discovery, the potential impact of AI to the business bottom line and beyond is staggering.
With interest and investment in AI at an all-time high, it is important to get beyond the hype to fully understand the capabilities and limitations of this technology and determine if and how it fits into your business strategy. Often, misinformation surrounding the implementation of AI leads to unrealistic expectations and puts the ROI of AI projects at risk. Here, I aim to dispel some common AI myths and highlight key considerations to drive success in AI implementation.
Myth 1: Artificial intelligence can make business decisions for you
The truth is that AI alone is not well suited for complex higher-level human tasks, though it can inform them. Today’s markets are extremely dynamic and influenced by a wide array of ever-evolving factors. The challenge is to accurately train an algorithm to fully account for such diversity. AI outcomes are also heavily influenced by the quality and depth of data inputs, meaning blind reliance on its conclusions can be risky.
Instead, AI is better deployed to dramatically improve workflow efficiencies for well-defined processes, freeing up human capacity for more complicated tasks. The power of AI is that it can enable high accuracy functions not possible with traditional programming. This allows far more possibilities to be considered and compared. AI is also able to recognize patterns in data and make recommendations based on these. For example, CAS is exploring AI to create smarter capabilities for content searches in our products by rapidly combining complex natural language processing and recommender systems.
Myth 2: Every business challenge can be helped by artificial intelligence
The truth is that AI is not an all-encompassing solution to inefficiency or productivity problems. Though early success can fuel a desire to apply AI generously across an organization, in reality, overly broad AI implementations that attempt to target many business challenges simultaneously often struggle to serve a diverse set of stakeholders, can’t live up to overblown expectations and fail to deliver meaningful ROI.
AI benefits from targeting singular rather than disparate problems. For example, AI can be effectively developed for the specific task of recognizing data on chemical structures. By a process of supervised learning, AI studies thousands of structures to become proficient at “recognizing” them, accounting for common shorthand, font or handwriting variations, etc.
So where should you start with AI implementation? Though others might disagree, it is my opinion that AI is best deployed to optimize and enhance the processes germane to your core competency first. Capitalizing on what differentiates your offering and drives your success, and extending your advantage there, maximizes the competitive impact AI can deliver.
Myth 3: Artificial intelligence can process any and all available data
Truth: Artificial intelligence engines are actually very picky eaters. One of the factors most often overlooked by organizations before starting an AI project is the content management and data modeling pre-work and on-going effort necessary to develop and maintain a clean, robust, uniform data set before starting an AI project.
AI engines are trained through a data-demanding iterative process. They become competent to perform a task efficiently through discovering the important features during the training process. Deep learning is a data-intensive process and pattern recognition requires comprehensive datasets to be executed accurately.
In the case of scientific and technical data, there are many aspects that present challenges. These include diverse naming conventions for chemical compounds and biologic entities, the use of images (graphs, charts, diagrams, etc.) rather than text to encode key information, varying units, etc. Many organizations also have historic information trapped in hand-written or other non-digital formats that must be digitized to be included in their dataset.
Though development of your dataset is far less exciting than the AI implementation, and can feel like a blocker to progress, it is a critical investment that is foundational to the success of any AI project. If patience is short, consider consulting external expertise on how to manage and model similar types of data to accelerate your project without sacrificing the desired outcome.
Interested in exploring the potential of AI for your business, but not sure where to start? Looking for a diverse corpus of human-curated scientific data to train an AI engine?
With more than 110 years of experience, no one knows more about managing and organizing scientific information than CAS. Lean on our content, technology and expertise to help make your AI vision a reality. Contact us to start the conversation today!