June 30, 2024
Kannan Subbiah
FCA | CISA | CGEIT | CCISO | GRC Consulting | Independent Director | Enterprise & Solution Architecture | Former Sr. VP & CTO of MF Utilities | BU Soft Tech | itTrident
AI’s “black box” problem is well-known, but the ethical imperative for transparency goes beyond just making algorithms understandable and its results explainable. It’s about ensuring that stakeholders can comprehend AI decisions, processes, and implications, guaranteeing they align with human values and expectations. Recent techniques, such as reinforcement learning from human feedback (RLHF) that aligns AI outcomes to human values and preferences, confirm that AI-based systems behave ethically. This means developing AI systems in which decisions are in accordance with human ethical considerations and can be explained in terms that are comprehensible to all stakeholders, not just the technically proficient. Explainability empowers individuals to challenge or correct erroneous outcomes and promotes fairness and justice. Together, transparency and explainability uphold ethical standards, enabling responsible AI deployment that respects privacy and prioritizes societal well-being. This approach promotes trust, and trust is the bedrock upon which sustainable AI ecosystems are built.Long
Organizations should ask themselves some serious, searching questions about why they are driven to keep doing the same thing over and over again – while spending millions of dollars in the process. As Bathurst put it: Why isn't security by design built in at the beginning of these projects, which are driving people to make the wrong decisions – decisions that nobody wants? Nobody wants to leave us open to attack. And nobody wants our national health infrastructure, ... But at this point, we should remind ourselves that, despite that valuable exercise, both the Ministry of Defence and the NHS have been hacked and/or subjected to ransomware attacks this year. In the first case, via a payroll system, which exposed personal data on thousands of staff, and in the second, via a private pathology lab. The latter incursion revealed patient blood-test data, leading to several NHS hospitals postponing operations and reverting to paper records. So, the lesson here is that, while security by design is essential for critical national infrastructure, resilience in the networked, cloud-enabled age must acknowledge that countless other systems, both upstream and downstream, feed into those critical ones.
“Customers are always going to have some challenges, and there are constant new technological trends evolving. Digital transformation is about intentionally moving towards making the experience more personalized by weaving new technology applications to solve customer challenges and deliver value,” shared Krishnan. However, as machine learning and GenAI help companies personalize their products and services, the tools themselves are also becoming more niche. “I think we’ll move to more domain and industry-specific generative AI and large language models. The healthcare industry will have an LLM, consumer packaged goods, education, etc,” shared Krishnan. “However, because companies will protect their own data, every large organization will create its own LLM with the private data. That’s why generative AI is interesting because it can actually get to be more personalized while also leveraging the broader knowledge. Eventually, we may all have our own individual GPTs.” ... Although new technologies such as GenAI and machine learning have had an immense impact in such a short time, Krishnan warns that guardrails are necessary, especially as our use of these tools becomes more essential.
领英推荐
Cognitive load is the amount of mental processing necessary for a developer to complete a task. Companies generally have one programming language that they use for everything. Their entire toolchain and talent pool is geared toward it for maximum productivity. On the other hand, CI/CD tools often have their own DSL. So, when developers want to alter the CI/CD configurations, they must get into this new rarely-used language. This becomes a time sink as well as causes a high cognitive load. One of the ways to avoid giving developers high cognitive load tasks without reason is to pick CI/CD tools that use a well-known language. For example, the data serialization language YAML — not always the most loved — is an industry standard that developers would know how to use. ... In software engineering, feedback loops can be measured by how quickly questions are answered. Troubleshooting issues within a CI/CD pipeline can be challenging for developers due to the need for more visibility and information. These processes often operate as black boxes, running on servers that developers may not have direct access to with software that is foreign to developers.?
"It starts by understanding how people with disabilities use your online platform," he said. While the accessibility issues faced by people who are blind receive considerable attention, it's crucial to address the full spectrum of disabilities that affect technology use, including auditory, cognitive, neurological, physical, speech, and visual disabilities, Henry added. ... The key is to review accessibility during content creation with a diverse group of people and address their feedback in iterations early and often. Bhowmick added that accessibility testing should always be run according to a structured testing script and mature testing methodologies to ensure reliable, reproducible, and sustainable test results. It is important to run accessibility testing during every stage of the software lifecycle: during design, before handing over the design to development, during development, and after development. A professional and thorough testing should take place before releasing the product to customers, Bhowmick said, and the test results should be made available in an accessibility conformance report (ACR) following the Voluntary Product Accessibility Template (VPAT) format.
Cloud-native practices, patterns, and technologies enhance the benefits of SaaS and COTS while reducing the inherent negatives by:Providing an extensible framework for adding new capabilities to commercial applications without having to customize the core product. Leveraging API and event-driven architecture to bypass the need for custom data integrations.?Still offloading the complexity of most infrastructure and security concerns to a provider while gaining additional flexibility in scale and resilience implementation.?Enabling opportunities to innovate core business systems with emerging technologies such as generative AI. Enterprises relying on SaaS or COTS still need the flexibility to meet their ever-evolving business requirements. As we have seen with advances in AI over the past year, change and opportunity can arrive quickly and without warning. Chances are that your organization is already on a journey to cloud-native maturity, so take advantage of this effort by implementing technologies and patterns, like leveraging event-driven architectures and serverless functions to extend your commercial applications rather than customizing or replacing them.