If you're a regular user of AI, either professionally or casually, you'll understand that it is far from the omniscient force that it is often depicted as.
Gen AI technologies – especially chatbots – are fallible things, despite the authoritative tone they often take. This manifests itself in errors, falsehoods or 'hallucinations', which are moments when AI seems to lose all reasoning.
The tricky thing is that it's sometimes hard to spot when AI has made a mistake, especially when it's dealing with complex data.
To combat this, lots of companies are investing in what is known as 'explainable' AI. This is a type of AI model that shows you the steps it has taken to reach its conclusions.
The hope is that, by showing its working, explainable AI (also known as XAI) is more transparent and easier to control.
In many ways, XAI can be viewed as an implementation of a social and legal "right to explanation", a principle that provides individuals with the ability to question and comprehend the automated decisions that impact their lives.
In this week's Top 10, AI Magazine spotlights some of the best and most impactful explainable AI tools currently available to both business users and casual users.
10. DataRobot's Transparent AI Platform
CEO: Jeremy Achin
HQ: Boston, Massachusetts, US
Specialisation: A comprehensive, automated machine learning (ML) platform that includes model explainability as a core feature.
What sets DataRobot apart is its 'all-in-one' philosophy. Rather than treating explainability as an afterthought, the firm has baked it right into its ML platform.
With DataRobot's systems, users can expect to get sophisticated predictive modelling capabilities that is totally transparent, allowing them to understand what's going on under the hood.
This integrated approach makes complex AI more accessible to a much broader audience, not just the data science whizzes and true blue tech afficionados.
9. Oracle's Cloud Infrastructure Data Science
CEO: Safra Catz and Executive Chairman Larry Ellison
HQ: Austin, Texas, US
Specialisation: Providing integrated AI governance and explainability features within its cloud infrastructure to help customers trace and understand how their models make predictions.
Oracle's approach to XAI is refreshingly holistic. Rather than just offering standalone tools, the company has tried to embed explainability into its broader AI strategy, which focuses a great deal of responsibility and compliance.
Oracle's Cloud Infrastructure Data Science platform can help users understand the model's thinking, coming with integrated ethics review boards and robust data privacy measures.
This is particularly appealing for industries where transparency is a must, like in the healthcare and finance sectors.
8. Nvidia's GPU-Accelerated SHAP
CEO: Jensen Huang
HQ: Santa Clara, California, US
Specialisation: Enabling the practical, commercial-scale deployment of computationally expensive explainability techniques like SHAP by leveraging the power of GPU acceleration.
Under CEO Jensen Huang's leadership, Nvidia has successfully evolved from its gaming origins to become an essential player in the modern AI market.
Methods like Shapley Additive exPlanations (SHAP) excel at understanding model decisions, but they're also computational monsters.
Nvidia's has been making these techniques more commercially viable by accelerating them on their GPUs.
The result? Financial institutions can now generate explainability profiles for entire portfolios in minutes rather than days. It's a game-changer for industries where time really is money.
7. Intel's Explainable AI Tools
CEO: Lip-Bu Tan
HQ: Santa Clara, California, US
Specialisation: An open-source toolkit for post-hoc model explainability and report generation, specifically designed to run best on Intel hardware.
Intel's explainable AI toolkit is a strategic alignment between software capabilities and hardware optimisation.
The firm's open-source toolkit includes a Model Card Generator and an Explainer module, but here's the twist – it's all optimised to run smoothly on Intel hardware.
It's strategic, but it's also rather sensible. By providing tools for fairness, interpretability and visualisation that perform best on their CPUs and GPUs, Intel's encouraging adoption of its chips within the AI ecosystem.
6. SAS's Viya
CEO: James Goodnight
HQ: Cary, North Carolina, US
Specialisation: A unified, governed platform with embedded explainability for trusted, auditable decisions in high-stakes industries like financial services and healthcare.
For SAS, explainability is a fundamental component of its AI technologies, particularly its Viya platform.
The team at SAS has decades of experience in enterprise analytics and it shows. The firm's AI tools are specifically tailored for regulated environments where compliance and accountability are of the utmost importance.
In this space, SAS has managed to tackle the thorny issue of explaining complex, autonomous AI systems that continuously adapt and make sequential decisions, providing a great example of how AI models will most likely look in years to come.
5. Amazon's SageMaker Clarify
CEO: Andy Jassy
HQ: Seattle, Washington, US
Specialisation: A feature within the AWS SageMaker platform that detects potential bias and explains model predictions at scale using model-agnostic methods like SHAP.
Amazon's journey from online bookstore to global tech behemoth is well-documented, but its contribution to explainable AI deserves special attention when looking at the history of the world's largest online retailer.
SageMaker Clarify is Amazon's answer to the growing demand for ethical AI. What sets it apart is its dual focus on bias detection and model explainability.
Clarify can spot potential data imbalances and help users to fix them, while also generating detailed explanations of model predictions.
And it works across tabular data, natural language processing and computer vision models – all at the massive scale that enterprise applications demand.
4. Salesforce's Einstein
CEO: Marc Benioff
HQ: San Francisco, California, US
Specialisation: An AI platform that integrates predictive analytics, NLP and intelligent recommendations across the entire CRM platform to automate tasks and deliver personalised experiences.
Salesforce Einstein represents something quite special in the XAI world. Rather than being a standalone tool, Einstein is woven right throughout the Salesforce product suite.
What this means is that business users get access to sophisticated AI without needing a computer science degree.
By applying AI across Sales Cloud, Marketing Cloud, Commerce Cloud and Service Cloud, Salesforce has made explainability part of the user experience rather than a separate technical requirement, making AI far more democratic in the process.
3. Microsoft's InterpretML
CEO: Satya Nadella
HQ: Redmond, Washington, US
Specialisation: An open-source package for training interpretable "glass box" models (e.g., Explainable Boosting Machine) and explaining complex "black box" systems, with a focus on debugging and fairness.
Microsoft was founded by Bill Gates and Paul Allen in 1975 and has since evolved from a software company into a cloud computing and AI powerhouse.
InterpretML's hybrid approach is particularly clever. The open-source package offers both inherently interpretable "glass box" models – such as their own Explainable Boosting Machine developed at Microsoft Research – and post-hoc "black box" explainers like LIME and SHAP.
In essence, it gives data scientists the flexibility to choose their weapon: build a transparent model from scratch or get insights into existing ones.
This dual-pronged strategy recognises that different problems require different solutions.
2. IBM's AI Explainability 360
CEO: Arvind Krishna
HQ: Armonk, New York, US
Specialisation: A comprehensive, open-source Python toolkit and taxonomy of diverse explainability methods for developers and researchers.
AI Explainability 360 is IBM's love letter to the developer community. It's a structured taxonomy designed to help users navigate the complex market of interpretation and explanation methods.
The toolkit supports popular methods like LIME and SHAP, but what's particularly useful here are the tutorials for real-world industrial use cases.
Users can look into situations like credit card approval, medical expenditure, proactive retention and much more.
It also represents IBM's attempt to bridge the gap between academic research and practical application.
1. Google's What-If Tool
CEO: Sundar Pichai
Company's HQ: Mountain View, California, US
Specialisation: An open-source, visual application for interactively probing and analysing ML models with a focus on fairness and hypothetical analysis.
What makes the What-If Tool particularly brilliant is its user-centric design. It doesn't require much coding at all, making it accessible to a much broader audience than most AI models.
The tool allows practitioners to test model performance in hypothetical situations, analyse feature importance and visualise model behaviour across multiple models and data subsets.
It's particularly strong when it comes to measuring systems according to multiple ML fairness metrics, which is a critical requirement for responsible AI development.
In a world where AI bias can have real-world consequences, tools like this will be hugely important going forward.
