Top 5 AI Trends to Look Out for in 2023

Over the past decade, the democratization of artificial intelligence (AI) in the enterprise space has accelerated. Balakrishna DR, EVP and head of AI and automation at Infosys, discusses the five key AI trends for enterprises to pay attention to in the near future.


Many organizations are increasingly maturing their AI pursuits. Interestingly, as AI is getting more embedded into the business landscape and new use cases emerge, some positive trends, I believe, now have the potential to take center stage in the coming year. Here’s a bird’s eye view, from my perspective.

1. AI for Text, Speech and Vision Will Continue its Journey into Mainstream 

There is a treasure trove of intelligence buried deep in the conversations that customers have with contact center executives. These unstructured voice and text-based conversations are fast becoming one of the easiest sources of intelligence. In some scenarios, it is possible to derive crucial consumer insights to improve products and services, design virtual assistants to help the staff tackle complex customer issues and improve customer satisfaction. Some other intelligence that can be of value is identifying frequently asked questions and creating proper self-service channels for them, increasing customer engagement and identifying and prescribing opportunities for cross-selling and upselling and an abundance of other allied opportunities. Also, language and accent neutralization capabilities can enable an executive to serve customers across geographies. 

There are several existing hurdles in building these solutions, like achieving clean transcription from different languages, diverse dialects and accents, identifying different types of contextual vocabularies,  removing ambient noise, and using different channels like mono or stereo for recording the conversations. Over the years, big technology players have come up with many solutions. They have built powerful proprietary models with very high accuracies. But the major challenge is that the data needs to be sent over the cloud, which may conflict with confidentiality and privacy concerns. Also, these proprietary models have limited scope in training for domain-specific customizations. 

What will be a differentiator in the coming days is the usage of powerful deep learning to build encoder decoder transformer networks using pre-trained components and transfer learning. These computationally intensive models are leveraging the hardware acceleration of high-performance GPU computing to circumvent challenges with translations and nuances of speech. 

Large language models like BERT, and GPT-3, will become far more sophisticated in the coming days extending their capabilities to process different semantic similarities and contextual relationships and improving upon existing applications in text summarization and generation, chatbots, increasing translation accuracy and enhancing sentiment mining, search, code generation etc. 

In the field of computer vision, newer and more powerful models for object detection, segmentation, tracking and counting are being built that deliver previously unimagined levels of accuracyAugmented by extraordinarily powerful GPU, these models will become increasingly commonplace. 

We can expect to see hybrid solutions leveraging all the above advancements to bring the next generation of AI assistants to life. These solutions will have the warm touch of human conversations coupled with the fast execution and inferencing capabilities, ultimately resulting in lower operational costs and huge boosts in customer satisfaction.

See More: Why You Should Apply Caution When Using AI in Code Development

2. Generative AI in the Art and Creative Space

Attracting and retaining the mindshare of your customer base is a challenge that most enterprises are constantly struggling with. To improve your brand recall, you need to constantly generate quality content that is relevant and engaging and properly appropriated for circulation in a variety of outlets. Here comes Generative AI, which offers new capabilities to augment content creation. Using generative AI, Enterprises can create a variety of content like images, videos, and written material and decrease turnaround time. Generative AI networks employ transfer-style learning or general adversarial networks to create immersive content from different sources. Apart from obvious use cases in marketing, it can potentially revolutionize the media industry. Film making and restoring old films in high definition, augmented capabilities for special effects and building avatars in the metaverse are a few limitless applications. 

Here, large language models like GPT-3 will again come into play to create engaging content in fiction, non-fiction and academic articles. On many publicly available websites, it is already possible to generate quality images of abstract ideas that are rendered from simple written prompts from the user. It is possible to create narrations and voices in thousands of tones and frequencies in areas like audio synthesis. One of the malicious applications that might arise that we need to be vigilant of is the creation of deepfakes (artificially generated fake images and videos) which will lead to emerging threats like the proliferation of fake news and furthering harmful propaganda. Techno Thus, Generative AI will be a major transformational force augmenting our innate creativity in various business pursuits. 

3. Explainable AI to Make Ethical and Responsible AI a Reality

Increasingly enterprises are realizing the need for explainable AI to improve transparency, establishing accountability and exposing biases in automated decision-making systems. Explainable AI is also a major instrument in mitigating the risks inherently associated with Enterprise AI. It is also proven that explainable AI also increases AI adoption across the enterprise as people are more comfortable when AI models give justifications and rationale along with their predictions. In settings like healthcare or financial services, this would gain a lot of momentum as you would need to understand and articulate the justification for recommending a treatment or diagnosis or why a loan application was rejected. 

Several techniques, like LIME, increase model interpretability by perturbing the inputs and assessing impacts on the output. Another popular technique, SHAP, uses a game theory-based approach by analyzing a combination of features and their corresponding effects on the resulting delta. It creates explainability scores for highlighting the input’s aspects that contributed more to the output. For example, in image-based predictions, the dominant area or the pixels that resulted in the output can be highlighted. As the impact of AI continues to surge in business and society, we also expose ourselves to various ethical issues that arise from these complex use cases. Proper data governance frameworks, tools to expose bias and factors in transparency are being looked at to stay compliant with legal and social structures. Models will be thoroughly tested for drifts, humility, and bias. Proper model validation and audit mechanisms with inbuilt explainability and checks for reproducibility will become the norm, for safeguarding against ethical lapses. 

4. Adaptive AI Sharpening and Elevating Customer and Brand Experiences

Leading Retailers are investing significantly in improving operational efficiencies and customer experiences through AI. Increasingly, retail stores will become the focal point for driving brand awareness and customer experience rather than simple transaction centers, and Adaptive AI will be the force behind this transformation. Frictionless shopping experiences built on computer vision and edge-based AI systems that reduce wait times and reduce hassle will be one major growth area. The future retail stores will also be able to offer hyper-personalized recommendations and craft seamless customer journeys based on real-time insights generated through video analytics powered by on-prem infrastructure. 

In-store analytics will give intelligent insights based on the dwell time across different aisles in a store. Integrating with past shopping histories across multiple channels and factoring in the demographic profile will enrich the customer experience and make experiential shopping highly immersive and enjoyable. Omnichannel management will be augmented through adaptive AI, which will give highly contextual assistance. Conversational AI, coupled with emerging technologies like AR and VR, will augment the capability of store employees to redefine the shopping experience entirely in brick-and-mortar stores. 

5. Edge AI Will Become More Ubiquitous 

Edge AI has immense power to transform our daily lives by making common consumer devices context-aware through powerful deep learning. Edge-based AI will get more affordable due to lighter models and accessibility to high-performance GPU compute. The Edge models use local context-based learning and synchronize with the central model at the appropriate times, resulting in lesser bandwidth and energy requirements. These affordable and intelligent devices will revolutionize various segments like retail, manufacturing and energy utilities for use-cases like quality inspection, predictive maintenance and health and safety. 

The drop in costs due to lower compute requirements will give rise to a market for smart and responsive devices. Lesser data requirements will be a boon for sectors like healthcare and finance, where data management is strictly regulated. In every edge device’s models are customized to the specific edge environment, and critical data never exits outside the edge network. Edge AI will become pervasive in arenas like smart warehouses, manufacturing and utilities. As enterprises become more aware of bulky models’ huge energy requirements, edge-based AI will be adopted to reduce AI’s carbon footprint and meet sustainability goals.