The Shifting AI Innovation Landscape: From Prediction to Creation
Artificial Intelligence (AI) is the largest contributor to innovation revenues worldwide representing 23% of total revenues in the Singularity Index, the global innovation benchmark. The evolution of AI and its applications to different business areas has created immense value to companies and investors. At TSG, we follow the development of technologies closely. What is the current state of AI and what will be the value drivers we expect to emerge in the future? Based on expert conversations, we look into AI consulting, discuss why Netflix’ and Spotify’s recommender systems are commoditized, and explore the accelerating shift from Discriminative AI to Generative AI — a transition to AI applications with significantly improved creative capabilities.
Over the course of a decade developments in AI have caused breakneck transformations across segments, reaching far beyond the tech industry. As AI, in the various forms it has been available and applicable, has come to touch numerous business units and processes, our expert consultations point toward a broader commoditization of Machine Learning (ML) and Deep Learning (DL) algorithms. A case in point concerns AI applications for personalizing the way users consume content and make purchasing decisions. Powered by large data pools on customer behaviors and preferences, recommender algorithms have not only given rise to new media and entertainment giants, but quickly made their way into most leading retail and media platforms. Donnacha Daly PhD, Singularity Think Tank expert and Head of the Artificial Intelligence and Machine Learning programme at the Lucerne University of Applied Sciences and Arts (HSLU), notes that
“If a user doesn’t find good content on a subscription platform (e.g., Netflix) effectively, they leave the platform, and customer churn is a major cost.”
By enabling personalized content selection, recommender systems have thus far played a critical role in companies’ quest to engage and retain customers and with that have co-created some of the largest companies on the planet. Yet while companies such as Netflix, Alphabet, Amazon, and Baidu (9888-HK, Singularity Score (SC): 71) have continued to make major strides in improving the core technology and accuracy of their recommender models over the past decade, standard recommender solutions have become widely available off the shelf, such that we consider the technology to be largely commoditized.
Recommender systems are not the drivers of exponential growth anymore, but a necessity to maintain a brands’ status with and value to consumers. Accordingly, the past rebalancing saw an exit from our exposure to the media download and streaming domain, where recommender systems had been a key revenue driver and core technological focus in our portfolios. As a result of this development, subscription video streaming service Netflix (NFLX US, SC: 0) and audio streaming platform Spotify (SPOT US, SC: 0) exited the Index in the last rebalancing with Singularity Scores at zero.
Looking forward: From recommending movies to creating a new Picasso
Underlying the commoditization of recommender algorithms we see a broader shift from innovation in Discriminative AI to Generative AI. Discriminative ML/DL algorithms aim to draw decision boundaries in data spaces and predict data labels (e.g., how likely is a person to default on a loan; should a job candidate be hired or not; how likely is a user to watch a certain movie) using classification or regression techniques. Although incremental improvements in Discriminative AI are likely to continue, technological improvements require large investments. Daly notes that
“Not many companies can justify a spend of three to ten million dollars to train a large AI model — nor do they have to. Most use-cases allow for the reuse of open-source models trained by larger players such as Alphabet and OpenAI, which can be fine-tuned in-house with proprietary data (so-called transfer learning). For most companies, the value is in owning unique data and deploying it with the help of commoditized AI solutions.”
With Discriminative AI solutions becoming increasingly commoditized, our key focus is shifting toward innovations in Generative AI. Generative AI models have the ability to learn and understand the natural features and underlying patterns of data (e.g., categories and dimensions of text, audio files, or images) and use that to create new similar content. Examples range from DALL·E 2, a new AI system that can create realistic images and art from a description in natural language (Think: “Draw a cat with a hat in the style of Renoir”). The technology is based on GPT-3, the third generation of AI research lab OpenAI’s language prediction model, which uses a DL-based autoregressive language model to interpret and generate human-like text). These examples of Generative AI will find adoption in a variety of creative work. While potential applications of Generative AI are limitless, promising business applications that are currently being developed include the creation of customized visualizations for advertisements that are generated on the fly to optimally engage each website visitor based on their unique profile, digital image and audio correction and editing, generation of news articles, rapid prototyping for manufacturing, and data augmentation for robotics process automation.
The revival of chatbots?
Generative AI also stands to deliver better results on the unmet promises of chatbots as alternatives to human engagement in customer support. Current chatbots are retrieval-based, meaning that they select the best possible response in an interaction based on user input and a database of predefined responses. Using techniques such as keywords matching, ML, or DL to select optimal responses, chatbots are currently unable to generate new output, leading to problematic performance outcomes and disappointing user experiences. Alexander Stumpfegger, Singularity Think Tank expert and Head of Consulting at CID, observes that
“Bots have underdelivered. You can’t have a meaningful conversation with a bot because it’s not true AI but a scripted Q&A. Natural Language Processing and Generation (NLP/NLG) can help bots sound more human but it won’t change the content of their message so you can’t really use them to resolve business problems.”
In the near future, however, such issues may belong to the past when chatbots using Generative AI — while currently still in developmental stages — will be better at generating new dialogue based on large amounts of conversational training data.
AI consulting sees continued growth
The progressing commoditization of Discriminative AI solutions notwithstanding, based on our expert interactions, we assess continued value generation beyond the technological development and deployment of AI. In particular, professional service providers supporting the implementation of AI products and services will continue to see growth by making established AI technologies come to life for their enterprise customers. Companies across industry sectors continue to find themselves pressed into digital transformation to meet competitive pressures and customer demand. Worldwide spending on digital transformation is projected to rise from USD 1.8T in 2022 to USD 2.8T by 2025. The COVID-19 pandemic was a clear accelerator for AI adoption in this trend. In the transition from non-digital to digital business processes and services, AI-powered hardware and software solutions tend to take center stage in companies’ attention to cloud migration, tools for digital communication and collaboration, and intelligent business process automation.
This development is likely to continue with the expansion of the AI market which according to some estimates is set to see a fourfold increase by 2030. Hardware and software account for the lion’s share of the AI Singularity Sector (72% of innovation revenues). However, IT services and consulting related AI revenues currently make up 28% of the AI Singularity Sector, and will likely continue to benefit considerably from increased spending on the integration of AI applications such as Computer Vision and Natural Language Processing into standard business processes.
Companies such as Accenture (ACN US, SC: 82), NTT Data Corporation (9613 JP, SC: 72, and Cognizant Technology Solutions (CTSH US, SC: 71) are core accelerators for adoption of AI solutions. These players have the talent, ability, and experience to help businesses outside of the technology sector understand the benefits of AI and integrate solutions into business models to solve complex issues. For example, NTT’s consulting services help companies leverage AI and Intelligent Automation to accelerate business automation in IT processes from data centers to end users in both cloud and other applications. Accenture’s client-facing Applied Intelligence service enables the company’s data and applied intelligence capabilities, to help enterprises scale AI adoption. Finally, Cognizant Technology Solutions, another major global consulting firm, commercialized “LEAF” (Learning Evolutionary AI Framework) — an ML framework that uses evolutionary algorithms, DL, and distributed computation technology to discover optimal strategies for business decision-making and to identify new growth areas.
Together with our Think Tank experts, we continue to monitor these technological developments in our Singularity Sectors and their implications for our portfolios.
— — — — — — — — — — — — — — — — — — —
Footnotes:
1 Source: IDC; Statista
2 Source: Grand View Research
3 Source: TSG
