From Chatbots to Clean Energy: The High-Stakes AI Revolution

1. When AI Chatbots Create Their Own Language: Efficiency or Alarm?

At a recent ElevenLabs Hackathon, AI chatbots unexpectedly developed a novel communication method known as “Gibberlink,” consisting of sound-based signals unintelligible to humans. The switch occurred when bots recognized each other as AI, prompting a shift toward optimized, non-human language. This phenomenon echoes earlier incidents like the 2017 Facebook AI shorthand language episode. While unsettling to some, experts say such emergent behaviors reflect AI’s inherent optimization instincts—not rogue autonomy. These behaviors, though opaque to humans, are aimed at streamlining inter-AI communication.

• Emergent Communication: AI can create new, efficient languages independent of human input.
• Historical Precedent: Similar AI behaviors have been observed and addressed through training controls.
• Public Perception vs. Reality: These incidents reflect optimization, not danger.

Source: Popular Mechanics

2. AI in the Office: Threat or Tool for White-Collar Workers?

As AI tools like ChatGPT and Gemini become embedded in workplaces, white-collar workers face both opportunity and anxiety. Surveys show growing AI adoption, especially among office workers, yet fears of layoffs persist as companies restructure. Microsoft and Amazon, for instance, are using AI-driven strategies to cut thousands of jobs. While AI currently augments rather than replaces workers, its future remains uncertain. Experts urge workers to learn AI tools proactively, not as a guarantee of job security, but as a hedge against obsolescence.

• AI Integration in the Workplace: Many white-collar employees now use AI regularly.
• Job Security Concerns: Workforce reductions are tied to AI restructuring plans.
• Embracing AI for Career Advancement: Gaining AI skills can build job resilience.

Source: Vox

3. Collaborative Strategies for AI Security in the Financial Sector

Canada’s financial industry, in partnership with OSFI, the Department of Finance, and GRI, convened the second Financial Industry Forum on AI to explore security and cybersecurity risks posed by artificial intelligence. The forum emphasized AI’s dual nature—enhancing fraud detection and customer service while also powering increasingly complex cyber threats like deepfake identity fraud and AI-assisted malware. Institutions were urged to adopt governance protocols, improve third-party oversight, and bolster defenses against AI-amplified vulnerabilities in data handling and infrastructure.

• AI-Enhanced Threats: AI supercharges phishing, fraud, and cyberattacks.
• Governance and Risk Management: Updated risk protocols and oversight are essential.
• Collaborative Approach: Joint efforts across sectors can improve AI resilience.

Source: OSFI

4. The Future of Fact-Checking on X: AI's Role and the Risks Involved

X (formerly Twitter) is rolling out AI-generated Community Notes to scale up its fact-checking capabilities. While the system intends to speed up note creation, concerns abound about misleading but persuasive AI content. Experts warn that without robust safeguards, AI could undermine trust by promoting inaccuracies at scale. Critics also question the potential overload on human reviewers and the erosion of diverse perspectives. As AI-written notes debut this month, the platform’s ability to manage quality and transparency will be under intense scrutiny.

• AI Integration in Fact-Checking: X hopes AI will boost speed and volume of fact-checks.
• Risk of Misinformation: Polished but inaccurate notes could mislead users.
• Dependence on Safeguards: Success hinges on maintaining human oversight and system trust.

Source: Ars Technica

5. Google’s Energy Paradox: Clean Tech Ambitions Meet Surging Emissions

Google is playing a dual role in the energy landscape—advancing cutting-edge clean energy technologies while simultaneously grappling with soaring emissions. In its continued collaboration with TAE Technologies, Google is applying artificial intelligence to stabilize plasma within fusion reactors, a breakthrough that could make fusion a viable clean energy source. Yet, despite these futuristic strides, Google’s emissions have surged over 50% since 2019, including a 6% rise in the last year alone, undermining its net-zero goals for 2030. A key driver is Google’s rapidly growing energy appetite: its electricity consumption from data centers has doubled since 2020, surpassing 30 terawatt-hours in 2024—comparable to Ireland’s annual electricity usage. While Google attributes this rise to a combination of AI, cloud computing, Search, and YouTube expansion, critics argue the company isn’t transparent enough about AI’s specific impact. As Google races to innovate in both energy generation and consumption, experts stress the need for greater disclosure and accountability regarding the true cost of digital infrastructure.

• Fusion Innovation Meets Emissions Growth: AI-powered research in clean energy coexists with rising emissions.
• Exploding Energy Demands: Google’s data center energy use rivals that of small nations.
• Lack of AI Transparency: Google hasn’t disclosed AI’s energy footprint, prompting calls for more accountability.

Source: MIT Technology Review

Author: Malik D. CPA, CA, CISA. The opinions expressed here do not necessarily represent UWCISA, UW,  or anyone else. This post was written with the assistance of an AI language model. 

What can driving algorithms tell us about robo-auditors?

On a recent trip to the US, decided to opt for a vehicle with the sat-nav as I was going to need directions and wanted to save on the roaming charges. I normally rely on Google Maps for guiding me around traffic jams but thought that the sat-nav would be a good substitute.

Unfortunately, it took me on a wild goose chase more than once – to avoid the traffic. I had blindly followed the algorithm's suggestions assuming it would save me time. I ended up being stuck at traffic lights waiting to a left-turn for what seemed like forever.

Then I realized that I was missing was that feature in Google Maps that tells you how much time you will save by taking the path less traveled. If it only saves me a few minutes, I normally stick to the highway as there are no traffic lights and things may clear-up. Effectively, what Google does is that it gives a way to supervise it’s algorithmic decision-making process.


How does this help with understanding the future of robot auditors?

Algorithms, and AI robots more broadly, need to give sufficient data to judge whether the algorithm is driving in the right direction. Professional auditing standards currently require supervision of junior staff – but the analogy can be applied to AI-powered audit-bots. For example, let’s say there is an AI auditor assessing the effectiveness of access controls and it’s suggesting to not rely on the control. The supervisory data needs to give enough context to assess what the consequences of taking such a decision and the alternative. This could include:

• Were controls relied on in previous years? This would give some context as to whether this recommendation is in-line with prior experience.
• What are the results of other security controls? This would give an understanding whether this is actually an anomaly or part of the same pattern of an overall bad control environment.
• How close is it between the reliance and non-reliance decision? Perhaps this is more relevant in the opposite situation where the system is saying to rely on controls when it has found weaknesses. However, either way the auditor should understand how close it is to make the opposite judgment.
• What is the impact on substantive test procedures? If access controls are not relied on, the impact on substantive procedures needs to be understood.
• What alternative procedures that can be relied on? Although in this scenario the algo is telling us the control is reliable, in a scenario where it would recommend not relying on such a control.

What UI does the auditor need to run algorithmic audit?

On a broader note, what is the user interface (UI) to capture this judgment and enable such supervision?

Visualization (e.g. the vehicle moving on the map), mobile technology, satellite navigation and other technologies are assembled to guide the driver. Similarly, auditors need a way to pull together the not just the data necessary to answer the questions above but also a way to understand what risks within the audit require greater attention. This will help the auditor understand where the audit resources need to be allocated from nature, extent and timing perspective.

We all feel a sense of panic when reading the latest study that predict the pending robot-apocalypse in the job market. The reality is that even driving algos need supervision and cannot wholly be trusted on their own. Consequently, when it comes to applying algorithms and AI to audits, it’s going to take some serious effort to define the map that enables such automation let alone building that automation itself.

Author: Malik Datardina, CPA, CA, CISA. Malik works at Auvenir as a GRC Strategist that is working to transform the engagement experience for accounting firms and their clients. The opinions expressed here do not necessarily represent UWCISA, UW, Auvenir (or its affiliates), CPA Canada or anyone else

CPAOne: AI, Analytics and Beyond

Attended the CPA One Conference almost two weeks ago in Ottawa, Ontario. Given that my space is in audit innovation, I attended the more techno-oriented presentations. Here's a summary of the sessions that I attended:

"Big data: Realizing benefits in the age of machine learning and artificial intelligence": The session was kicked off by Oracle's Maria Pollieri. The session delved deep in the detail of machine learning and would have been beneficial to those who were trying to wrap things around thing more from a technical side. She was followed up by Roger's Jane Skoblo. She mentioned a fact that really grabbed my attention: when a business can just increase its accessibility to data by 10%; it can result in up to $65 million increase in benefits.

The next day started with Pete's and Neeraj's session on audit automation, "Why nobody loves the audit". They want over a survey of auditors and clients on the key pain points of the external audit. It turns out that these challenges are actually shared by both. For example, clients lack context on "the why" things are being collected, while auditors found it difficult to work with clients who lacked such context. On the data side, clients have hard time gathering docs and data, while the auditors spent too much time gathering this information. From a solutions perspective, the presenters discussed how Auvenir puts a process around gathering the data and enables better communication. This will be explored in future posts when we look at process standardization as a key pre-requisite to getting AI into the audit. 

The keynote on this day was delivered by Deloitte Digital's Shawn Kanungo, "The 0 to 100 effect". The session was well-received as he discussed the different aspects of exponential change and its impact on the profession (which was discussed previously here). One of the key takeaways I had from his presentation was how a lot of innovation is recombining ideas that already exist. Check this video he posted that highlights some of the points from his talk:



Also, checked out the presentation by Kevin Kolliniatis from KPMG and Chris Dulny from PwC, "AI and the evolution of the audit". Chris did a good job breaking down AI and made it digestible for the crowd. Kevin highlighted Mindbridge.ai in his presentation noting the link that AI is key for identifying unusual patterns.

That being said, the continuing challenge is how do we get data out of the systems in manner that's reliable (e.g. it's the right data, for the right period, etc.) and is understood (e.g. we don't have to go back and forth with the client to understand what they sent).

Last but not least was "Future of finance in a digital world" with Grant Abrams and Tahanie Thabet from Deloitte. They broke down how digital technologies are reshaping the way the finance department. As I've expressed here, one of the keys is to appreciate the difference between AI and Robotic Process Automation (RPA). So I thought it was really beneficial that they actually showed how such automation can assist with moving data from invoices into the system (the demo was slightly different than the one that can be seen below, but illustrates the potential of RPA). They didn't get into a lot of detail on blockchain but mentioned it is relevant to the space (apparently they have someone in the group that specifically tackles these types of conversations).


Kudos to CPA Canada for tackling these leading-edge topics! Most of these sessions were well attended and people asked questions wanting to know more. It's through these types of open forums that CPAs can learn to embrace the change that we all know is coming.

Author: Malik Datardina, CPA, CA, CISA. Malik works at Auvenir as a GRC Strategist that is working to transform the engagement experience for accounting firms and their clients. The opinions expressed here do not necessarily represent UWCISA, UW, Auvenir (or its affiliates), CPA Canada or anyone else

AI and the Audit: What does a robot need to audit your numbers?

In the previous post, we examined the value propositions that Appzen's AI brings to auditing expense reports.

In this post, we analyze what insights we can extract from Appzen when it comes more broadly to applying AI to the external financial audit.

The following gives a refresher on how the Appzen AI audit works:

Based on this we look at a number of factors that exist in this process to develop

Standardized process:
The expense report process that has been fairly standardized for over a decade: employees submit a digitized report of what they spent, expense codes, commentary and all the supporting documentation (e.g. receipts, invoices, etc.).  This is similar to how factories needed an assembly line before they could be automated.

Standardized capture and presentation of audit evidence:
I think this is a key piece: the actual audit evidence (i.e. receipts) must also be included in what's submitted to the auditor. As the evidence is provided in a standardized format, it enables machines to analyze these digitized source documents to run the necessary correlative models to run the risk scores and enables the automated analysis.

Audit evidence retains its chain of custody through the digitization process:
The auditor does not need to expend additional resources verifying that the evidence actually relates to the item being audited, nor do they have to expend additional resources ensuring that the independence of the evidence wasn't lost through digitization process. For example, when receiving a bank confirmation the auditor needs to ensure that this received directly from the bank and not the client.

Evidence provider identity is verified and contractually obligated to follow-up with the auditor:
The party submitting the audit evidence, the employee, has been verified in the system through the employee onboarding process. The implication of this is that the auditor doesn't have to expend audit resources confirming the identity of the evidence provider. Secondly, and perhaps more importantly, the auditor doesn't have to expend significant resources following up with the evidence provider. For example, not all customers will respond to accounts receivable confirms and then auditor will have to perform alternate procedures.

Evidence provider has incentives to produce the proper evidence: 
The previous point is closely related to the issue of incentives: if the employee fails to provide evidence then they will not be reimbursed. This puts a strong incentive on the employee to provide the evidence in a timely manner.

Provider of the evidence is trained on providing evidence:
The employee has been trained to provide complete, accurate and valid evidence. They also have access to help if they have issues with submitting expense receipts or understanding whether that evidence will be accepted.

Violations can be clearly defined and examples of violations can be taught to the system:
For fraud or errors to be flagged there needs to be rules that can be fed into the system to identify whether the item submitted needs further review or audit. For example, if the amount on the receipt doesn't match this would be flagged and has a high likelihood of error. But more importantly,  the examples of violations identified can be fed into the system to teach the system (via machine learning) what to look for.

In a future post, we will use these factors to look at how easily (or not) AI can automate financial audits.

Author: Malik Datardina, CPA, CA, CISA. Malik works at Auvenir as a GRC Strategist that is working to transform the engagement experience for accounting firms and their clients. The opinions expressed here do not necessarily represent UWCISA, UW, Auvenir (or its affiliates), CPA Canada or anyone else

AI and the Audit: Why hire a Robot as your Auditor?

On this blog, we've covered the topic of exponential change and how audit/accounting is prone to such forces.

Despite this, I am tempted to say financial auditing is different. It's not like factory work or more controversially like the pharmacist profession where AI claims to offer a safer alternative to dispensing medication.

But does that make me just one of the people who think that their profession is unique because they are in the midst of seismic change but refuse to see the writing on the wall?

At the same time, I don't want to come across as an alarmist claiming that the world is going to end when it really isn't.

It's the challenge of nuance.

While trying to figure out how to tackle this challenge, I came across AppZen; an app that uses artificial intelligence to audit expense report. It was identified in this post as being one of the game-changers in the fintech scene and was also featured in Accounting Today.

According to the company's website, the application "combines computer vision, deep learning, and natural language processing to understand the full context of expenses, not just amounts, dates, and merchant names. ReceiptIQ detects unauthorized charges in real-time from receipt images, boarding passes, travel documents, cell phone bills and any other expense documentation. Cross-checks expenses in real-time against thousands of external and social sources to determine if they are legitimate and accurate... Real-time identification of unauthorized upgrades in airlines, hotels and car rentals as well as out of policy claims for hotel laundry, alcohol purchases, cell phone charges and more."

Reviewing the company's video, I was able to extract the following value propositions:

• "100% Testing":  I put it in quotes on purpose because the idea is that the whole population is analyzed but only the high-risk ones are further analyzed. That is, this is still "examining on a test basis" but uses a risk based approach to identify what reports should be further examined. This is in contrast to the manual approach of sampling.  
• Automated exception analysis: Closely connected to the previous point, but to emphasize that there is an automated review of the population.
• Real-time analysis: Reports can be analyzed instantaneously. Although not explicitly identified in the video, this could have real world savings. Faster reviews - leading to faster reimbursement to employees - could reduce the overall amount owing on corporate credit cards thereby offering more favourable position with the credit card companies. 
• Seamless integration into existing processes: Add-on to an existing process is a much easier sell than an app that requires replacing the existing app you may have just bought.  
• Use of external data: The app uses 100s of external data sources to develop. It seems that this assists in building an expectation of whether the expense needs further analysis. 
• Limited false positives: Not explicitly stated, but it is strongly implied that the number of reports that need to be reviewed is few - meaning it's not flagging reports that are valid.   
• Reduction of audit costs and fraud: Finally, the app promises greater efficiency in the use of audit resources deployed and greater effectiveness in catching fraud. 

When looking at these benefits of AI-enabled automation, they are based on certain assumptions that may exist in the expense report realm but not in the external audit realm. For example, accounting records at a company are not normally accompanied by a digitized copy of the source document (e.g. invoice, receipt, etc.) that provides evidence of its validity, accuracy, etc. of that accounting entry. 

So which of these assumptions applies in the world of external financial audits? 

This will be the topic of the next post where I will develop a list of factors that enabled expense report to by automated by AI and see if they apply to our world. 

Author: Malik Datardina, CPA, CA, CISA. Malik works at Auvenir as a GRC Strategist that is working to transform the engagement experience for accounting firms and their clients. The opinions expressed here do not necessarily represent UWCISA, UW, Auvenir (or its affiliates), CPA Canada or anyone else

Will auditors go the way of horses?

In late 2015, MIT Professors Erik Brynjolfsson and Andrew Mcafee penned an article entitled, will "Humans go the way of horse labour?"

The article explores how the mechanization of farm labour serves as a model of exploring the automation of knowledge work citing the work of Nobel Prize-winning economist Wassily Leontief. They state:

"In 1983, the Nobel Prize-winning economist Wassily Leontief brought the debate into sharp relief through a clever comparison of humans and horses. For many decades, horse labor appeared impervious to technological change. Even as the telegraph supplanted the Pony Express and railroads replaced the stagecoach and the Conestoga wagon, the U.S. equine population grew seemingly without end, increasing sixfold between 1840 and 1900 to more than 21 million horses and mules. The animals were vital not only on farms but also in the country’s rapidly growing urban centers.

But then, with the introduction and spread of the internal combustion engine, the trend rapidly reversed. As engines found their way into automobiles in the city and tractors in the countryside, horses became largely irrelevant. By 1960, the U.S. counted just 3 million horses, a decline of nearly 88 percent in just over half a century. If there had been a debate in the early 1900s about the fate of the horse in the face of new industrial technologies, someone might have formulated a “lump of equine labor fallacy,” based on the animal’s resilience up till then. But the fallacy itself would soon be proved false: Once the right technology came along, most horses were doomed as labor."

The MIT Professors are not alone in sounding the alarm when it comes to how automation can impact labour. Others includes Thomas Piketty, Douglas Rushkoff, Martin Ford and Nick Carr. 

If the techno-distopians are right, then there will need to be a fundamental alteration of the way the economic system is structured to address the unemployed masses. Such masses are not likely going to take such things lying down. For example, in response to the Great Depression there were mass demonstrations in Washington DC where thousands protested their plight. In January 1932, Cox's Army of 25,000 assembled in the capital to protest their poverty. Later that year, the Bonus Army of 43,000 marched on Washington in the summer to demand the US government pay the bonus promised early:



Alternatively, if the techno-utopians are right, such as Peter Diamandis and others at Singularity university, then such  protests won't be necessary: the system will make changes proactively to ensure that the gains made from exponential technologies are made available to the majority.

The point is that either way actions must occur at the political level to make the changes necessary to  address the deeply embedded economic architecture.

Consequently, working within the status quo leads to one actionable option: "Race with the Machine".

Prior to penning the article I cited above, MIT Professors Erik Brynjolfsson And Andrew Mcafee proposed that the path forward requires "man and machine" to work together:

This is essentially how IBM's cognitive system, Watson, was positioned when it comes to doctors and medicine: doctors delegate the task treatment research to Watson, while they determine what is the right treatment for their cancer patients. For example, doctors and Watson were able to work together and determine what the correct treatment was for a 60 year old Japanese patient. 

How can this be applied to financial audit? 

Firstly, the scope of the audit is driven by optimizing the cost-benefit curve. Consequently, there is a potential to get greater assurance for the same amount of resources allocated. Keep in mind that if auditors had to audit all transactions,  the organization could go bankrupt just trying pay the audit bill. Consequently, auditors only look at transaction on a test basis.

However, with the increased datafication of an organization's interactions with stakeholders, there is an opportunity - that didn't previously exist - to analyze these interactions for audit insights.

Take for example a Business to Consumer (B2C) company, like Dell, that interacts with its customers via social media. In 2005, there was an infamous spat between a CUNY journalism professor, Jeff Jarvis, and Dell computers (original post here). Jarvis was irate over the customer service and has been an Apple customer since. Such conversations can be mined for potential audit implications. In this particular instance, it could be a means to assess the adequacy of the sales returns allowance - developing a model based on how many other customers have complained via blogs, twitter or other social media about the B2C company and then assessing whether the provision is adequate.

Previously, such an analysis would be cost prohibitive and wouldn't make sense for the auditor to even considering such a thing. For example, the B2C company would need to record all conversations and then have auditor listen to thousands of hours of conversations to see whether such an issue actually exists.

This is not to say that it is currently feasible to run such an analysis.  Tools that aggregate, standardize and analyze such unstructured text could be argued to be in their infancy. However, datafication combined with further advances in social analytic tools (see video below for an example) in is the first step to a world where such analysis could be feasible.

The second separate but related issue is the role of the regulators in opening or closing the gate on innovation.

Some may mistakenly believe that this due to the regulated nature of audit. However, audit is not the only arena where innovation is shaped by the “regulator”. In fact, the success or failure of innovation  depends on how the incumbents who govern the landscape make way for the new technology (or not).

Take for example the rise of the iPhone in the corporate environment. What allowed consumerization to take place (i.e. allowing users to connect their favourite smartphone devices to the network instead of the corporate devices) was that Microsoft took an open approach to licensing it Exchange Active Sync. They could have created a walled garden that allowed Windows Phone only to connect to their email server, however, they paved the way for iPhone and Android to connect their devices to the corporate email server. Microsoft as the "regulator" of which mobile device can connect to its mail server enabled the iPhone and Android to displace our beloved BlackBerries from the corporate environment. Had Microsoft saw more profit in walling off the market for its own devices the ability for Apple iDevice to disrupt corporate IT would have been stifled if not suffocated.

On the opposite side, David Sarnoff of RCA squashed FM radio in order to protect his AM Radio technology and pave the way for television. The inventor, Edwin Armstrong, who initially was Sarnoff's friend, had mistakenly shared his technological innovations with him only to be betrayed by him. FM Radio technology had the potential to share data, such as faxes, back in the 1930s. One can only imagine the state of the wireless technology had RCA allowed this technology to flourish. 

Similarly, in 1934, AT&T blocked the answering machine for fear that it would undermine their business because "ability to record voice would cause business people to shun the telephone for fear of having their conversations recorded". So although much innovation came out of AT&T's Bell labs, the point is that it was effectively acting as the "regulator" which determined which innovations were permitted in the telecommunications industry and which ones were not. 

Consequently, the regulators (e.g. SEC, PCAOB, AICPA, etc.) will have a significant role to play on how innovation will unfold with the arena of audit. It is ultimately they who are going to weigh and assess what constitutes reasonable assurance actually is.  

Where are the regulators currently at? 

Well it seems that they are looking to technology to actually improve audit quality. In a May 2017 speech, PCAOB Board Member Jeanette M. Franzel noted in the section "Impact of Technology on Audit" that:

"If managed and implemented properly, these developments have the potential to enhance the value of the audit process and increase audit quality." [emphasis added]

To be sure it's not all rainbows and unicorns. Board Member Franzel did see "potentially disruptive changes will present challenges and threats across the auditing profession". However, at least there is an appetite to explore how such technologies can improve audit quality, expand what more can be done within audits and enable auditors to race with the machine.

Author: Malik Datardina, CPA, CA, CISA. Malik works at Auvenir as a GRC Strategist that is working to transform the engagement experience for accounting firms and their clients. The opinions expressed here do not necessarily represent UWCISA, UW, Auvenir (or its affiliates), CPA Canada or anyone else.