In Augmented Apps, we examine how product teams are exploring AI and Machine Learning to make their products more intuitive and enhance the user experience.
Artificial intelligence is transforming products in surprising and ingenious ways. Whether it’s core to the product, as with a stock market forecasting algorithm in Quants, or a peripheral component, such as a healthcare domain chatbot that diagnoses diseases via dialog with a patient, building reliable AI components into products is now part of the learning curve that product teams have to manage.
Certain proven methods and use cases are invaluable in helping product teams understand how to implement AI in apps to reproduce and enhance existing successes. In the case of a stock trading AI, for example, product managers are now aware that the data required for the AI algorithm must include human emotion training data for sentiment analysis. It turns out that emotional reaction is an important variable in stock market behavior!
Developing the right data model for your in-app AI is now a critical branch of programming, because correctly prepared data is vital to the algorithm model. It follows then that data scientists are suddenly integral to building embedded AI components.
A few AI components that have already achieved a high level of performance include X-ray and symptom diagnosis in healthcare, human emotion pattern recognition in stock trading apps, and innovative virus and malware detection in software systems. We’ll explore many of these use cases to build an overview of the state of the art in AI components today, starting with innovation in antivirus software.
Safeguarding software and users: AI in cybersecurity apps
One of the most exciting AI use cases today is happening in the field of cybersecurity, where AI methods are now commonly used to detect malware and malicious virus code embedded in files before they can do harm to systems. In fact, a breakthrough in AI security research occurred last year when researchers successfully used self-organizing incremental neural networks (SOINN) to identify malware and virus code.
The SOINN method converts binary files to visual representations and in doing so has achieved over 94% accuracy in detecting files infected with viruses! SOINN could become a standard use of AI for detecting viruses in security apps. By using a visual representation of code, the virus code can be detected without running the code and endangering the test system. Previously, machine learning-based malware detection algorithms were problematic because, according to Baptista and Shiaeles:
“…they rely on a virtual environment to analyze samples; this not only affects their runtime performance, but also endangers the whole system since the samples need to be executed. In addition, the ability of malware to adapt its behavior to the execution environment, the excessive number of features that need to be extracted per sample, and the high volume of malware instances being reported, reduce the chances of accurate detection.”
Breakthroughs like this — in which the fundamental representation of the subject is transformed — will certainly reshape and expand our concept of AI while providing new benefits.
Natural Language Processing: Teaching AI to talk
Natural language processing (NLP) unifies many of the other branches of AI through a human-oriented, AI-based interface. NLP performance improves significantly when the app is trained in a specific expert knowledge domain, such as symptom diagnosis. This is because a specialized context narrows the scope of the training data used in training the dialog algorithm. An immediately related branch of data science, called natural language querying (NLQ), focuses on using NLP to achieve particular goals.
Credit where it’s due: Understanding AI credit agents
A special use case of NLP and NLQ includes helping lending agents evaluate the creditworthiness of potential borrowers who have no prior credit history — the traditional basis of lending decisions. NLP can now mine vast and diverse data stores using trained patterns on alternative data sources that can help indicate the probability of reliable loan repayment.
For example, most credit applicants use smartphones. An AI can check their internet browsing patterns and see how closely they align with those of reliable clients. This type of supervised learning requires substantial data preparation: Important factors for correlation algorithms include a potential borrower’s social media activities, geolocation data, blogging contributions, peer networks, and relationship strength and duration. In fact, training metrics for these creditworthiness algorithms may bank on thousands of variables to generate an alternative credit score and also predict its own accuracy.
Predictive analytics AI boosts web app performance
Extraordinary advances in diverse applications such as visitor and log analytics owe their success to AI- and machine learning (ML)-based pattern recognition. Let’s look at a few such apps and their AI components and discover how we can benefit from their use cases:
- Web app visitor analytics: BI insights are derived from ML analysis of visitor stats and metrics and displayed via data visualizations
- Log analytics: The effectiveness of traditional error logs is enhanced by detecting app performance issues and by intelligently alerting targeted endpoints via email and Slack to critical user concerns
- Automated testing: Discovers bugs in new app releases using computer vision to determine page-rendering completeness
The potential uses of app behavior and visitor activity data stores are bounded only by the ingenuity of the data engineer. Analytics and BI platforms like Sisense can instantly provide data visualizations and alerts on visitor activity issues. Loggly and Timber likewise elevate traditional error logging by recognizing router issues as well as deployment and release failures. SmartBear and Tricentis enhance Selenium to provide AI-driven app testing automation. There is truly no limit to what can be achieved with a creative idea applied to data!
In all areas of predictive analytics, AI components perform best when the AI model can update its neural network during real-time transactions. This means feeding the client’s current preferences to the training data on the fly. AI web apps such as recommenders are the most obvious examples of real-time supervised training models that update themselves with each new transaction so that users can actually see the system learning in progress. And AI-based recommenders that update their own neural nets per transaction offer a more alive feeling in the user experience with such apps as:
- Swype personalized spelling suggestions (NLP)
- Ride-hailing apps’ suggesting destinations and best routes (geolocation)
- “Similar to favorite” music recommenders like Spotify
Healthcare benefits from AI diagnostics
As mentioned, NLP-powered chatbots with domain-specific expert knowledge can diagnose illness after a Q&A dialog with a patient. The next level in diagnostics is an AI-powered app that also takes a tissue sample from the patient and uses the lab output in conjunction with medical records and dialog to diagnose diseases, including cancer.
This amazing biotech application of AI uses genome sequencing in conjunction with neural nets to diagnose disease, outstripping the capabilities of traditional diagnosis methods. The app takes a patient’s tissue sample as input and then compares it with a database of every known pathogen (over 37,000 sequenced genomes!). The outcome is a comprehensive diagnostic report of possible disease agents.
Securing aviation hubs with AI
For those of us who are frustrated with long security lines at airport gates, new AI applications that automate the recognition of prohibited items in carry-on luggage will come as a refreshing change to speed up a sluggish human procedure. AI is also expanding rapidly on the frontier of emotion recognition and sentiment analysis. This will change the scope of preventive security to include biometrics of passengers who pose potential threats.
AI in apps: The future is today
AI apps are the future. On every horizon, we see automated systems ready to be dedicated to our safety, security, and comfort. At the moment, these systems are insulated from each other because data preparation for AI apps is a wild frontier, and there are no mutually intelligible standards for AI systems to share outcomes with each other.
However, just as the transmission control data protocol evolved toward a standard, so will data-wrangling eventually evolve standards for AI systems to learn from each other instead of exclusive datasets. When this happens, we will witness an exponential enhancement in the capability of AI now in its infancy. It’s an exciting frontier promising to deliver breathtaking benefits for everyone. Keep checking in as we bring you more insights from the emerging world of AI.
Inna Tokarev-Sela, Sisense’s Head of AI Research, has over 15 years’ experience in the tech industry. She spent the last decade at SAP, driving innovations in cloud architecture, in-memory products, and machine learning video analytics. A frequent speaker at industry events like IBC, NAB, Wonderland AI, and Media Festival, Inna holds a BS in physics and computer science, an MBA, and an MS in information systems, having written her thesis on neural networks.